A local inference system only wins under a workload, a utilization rate, and a performance
profile. Explore the crossover point against hosted API spend — with every assumption in view.
01 · model a scenario
Calculator
Start from a curated scenario or build your own. Results recompute locally, are shareable as a
URL, and can be saved in this browser.
A team runs long-context coding agents all day. Does a single dense node beat frontier API spend?At rack scale with the model's recommended topology, how far does the crossover move?Against the cheapest credible API, does owning hardware still make sense at all?Document-scale prompts with short answers shift the bottleneck from decode to prefill.The same purchase, derated to what a team on stock open-source engines is likely to reach.A team runs long-context coding agents all day. Does a single dense node beat frontier API spend?
Give the scenario a name before saving it.
Saved "WWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWW" in this browser.
Listed for this tab only: the browser refused to store it (a private window, blocked site data or a full quota). Use Share to keep its URL.
Recomputed locally · no data leaves this pageRecomputed locally · no data leaves this pageStale — resolve the findings below to update
Break-even duration
10mo
from acquisition
Net position at horizon
$1,168,481
after $420,000 all-in acquisition
Effective cost / M tokens
$1.79
API equivalent $5.11
Request throughput
535.7 / hr
decode-bound · 10,713,600 input tokens / hr
Break-even utilization
21%
you assumed 60%
Throughput evidence
medium
taken at face value
estimated
At 60% utilization DGX B300 costs less than Anthropic Claude Opus 5.5 after 10mo, and is $1,168,481 ahead by the end of the 3-year horizon — if your deployment reaches the modeled throughput, which is estimated evidence at medium confidence.At 100% utilization this never pays for itself: the hourly cost of running DGX B300 exceeds what the same work would cost on Anthropic Claude Opus 5.5. Over 19.99 years you would be $999,999,999,999,999 worse off.At 100% utilization and 2.00× quality parity the crossover against Anthropic Claude Opus 5.5 falls beyond the 19.99-year horizon — 999999y 11mo from purchase — so within the horizon you are $999,999,999,999,999 worse off. In cash it pays for itself after 19y 11mo and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity the crossover against Anthropic Claude Opus 5.5 falls beyond the 19.99-year horizon — 999999y 11mo from purchase — so within the horizon you are $999,999,999,999,999 worse off. In cash it pays for itself after 59d and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity the crossover against Anthropic Claude Opus 5.5 falls beyond the 19.99-year horizon — 999999y 11mo from purchase — so within the horizon you are $999,999,999,999,999 worse off. In cash it pays for itself after 23h 59m and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity the crossover against Anthropic Claude Opus 5.5 falls beyond the 19.99-year horizon — 999999y 11mo from purchase — so within the horizon you are $999,999,999,999,999 worse off. In cash it pays for itself only after 999999y 11mo: $999,999,999,999,999 worse off within the horizon.At 100% utilization and 2.00× quality parity the crossover against Anthropic Claude Opus 5.5 falls beyond the 19.99-year horizon — 999999y 11mo from purchase — so within the horizon you are $999,999,999,999,999 worse off. In cash it pays for itself only after 999999y: $999,999,999,999,999 worse off within the horizon.At 100% utilization and 2.00× quality parity the crossover against Anthropic Claude Opus 5.5 falls beyond the 19.99-year horizon — 999999y 11mo from purchase — so within the horizon you are $999,999,999,999,999 worse off. In cash it never pays for itself: $999,999,999,999,999 worse off over 19.99 years.At 100% utilization and 2.00× quality parity this never pays for itself: the hourly cost of running DGX B300 exceeds what the same work would cost on Anthropic Claude Opus 5.5, times that parity. Over 19.99 years you would be $999,999,999,999,999 worse off. In cash it pays for itself after 19y 11mo and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity this never pays for itself: the hourly cost of running DGX B300 exceeds what the same work would cost on Anthropic Claude Opus 5.5, times that parity. Over 19.99 years you would be $999,999,999,999,999 worse off. In cash it pays for itself after 59d and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity this never pays for itself: the hourly cost of running DGX B300 exceeds what the same work would cost on Anthropic Claude Opus 5.5, times that parity. Over 19.99 years you would be $999,999,999,999,999 worse off. In cash it pays for itself after 23h 59m and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity this never pays for itself: the hourly cost of running DGX B300 exceeds what the same work would cost on Anthropic Claude Opus 5.5, times that parity. Over 19.99 years you would be $999,999,999,999,999 worse off. In cash it pays for itself only after 999999y 11mo: $999,999,999,999,999 worse off within the horizon.At 100% utilization and 2.00× quality parity this never pays for itself: the hourly cost of running DGX B300 exceeds what the same work would cost on Anthropic Claude Opus 5.5, times that parity. Over 19.99 years you would be $999,999,999,999,999 worse off. In cash it pays for itself only after 999999y: $999,999,999,999,999 worse off within the horizon.At 100% utilization and 2.00× quality parity this never pays for itself: the hourly cost of running DGX B300 exceeds what the same work would cost on Anthropic Claude Opus 5.5, times that parity. Over 19.99 years you would be $999,999,999,999,999 worse off. In cash it never pays for itself: $999,999,999,999,999 worse off over 19.99 years.At 100% utilization DGX B300 costs less than Anthropic Claude Opus 5.5 after 19y 11mo, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — but a caveat below contradicts the throughput this rests on. Until that is resolved, treat the payback as the earliest it could possibly arrive, not a forecast: at one token per step the rate is overstated, so the savings are too, and the real payback is longer than the figure above.At 100% utilization DGX B300 costs less than Anthropic Claude Opus 5.5 after 19y, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — but a caveat below contradicts the throughput this rests on. Until that is resolved, treat the payback as the earliest it could possibly arrive, not a forecast: at one token per step the rate is overstated, so the savings are too, and the real payback is longer than the figure above.At 100% utilization DGX B300 costs less than Anthropic Claude Opus 5.5 after 59d, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — but a caveat below contradicts the throughput this rests on. Until that is resolved, treat the payback as the earliest it could possibly arrive, not a forecast: at one token per step the rate is overstated, so the savings are too, and the real payback is longer than the figure above.At 100% utilization DGX B300 costs less than Anthropic Claude Opus 5.5 after 23h 59m, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — but a caveat below contradicts the throughput this rests on. Until that is resolved, treat the payback as the earliest it could possibly arrive, not a forecast: at one token per step the rate is overstated, so the savings are too, and the real payback is longer than the figure above.At 100% utilization DGX B300 costs less than Anthropic Claude Opus 5.5 after 23h, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — but a caveat below contradicts the throughput this rests on. Until that is resolved, treat the payback as the earliest it could possibly arrive, not a forecast: at one token per step the rate is overstated, so the savings are too, and the real payback is longer than the figure above.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 19y 11mo, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — but a caveat below contradicts the throughput this rests on. Until that is resolved, treat the payback as the earliest it could possibly arrive, not a forecast: at one token per step the rate is overstated, so is everything computed from it, and the real payback is longer than the figure above. In cash it pays for itself after 19y 11mo and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 19y 11mo, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — but a caveat below contradicts the throughput this rests on. Until that is resolved, treat the payback as the earliest it could possibly arrive, not a forecast: at one token per step the rate is overstated, so is everything computed from it, and the real payback is longer than the figure above. In cash it pays for itself after 59d and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 19y 11mo, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — but a caveat below contradicts the throughput this rests on. Until that is resolved, treat the payback as the earliest it could possibly arrive, not a forecast: at one token per step the rate is overstated, so is everything computed from it, and the real payback is longer than the figure above. In cash it pays for itself after 23h 59m and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 19y 11mo, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — but a caveat below contradicts the throughput this rests on. Until that is resolved, treat the payback as the earliest it could possibly arrive, not a forecast: at one token per step the rate is overstated, so is everything computed from it, and the real payback is longer than the figure above. In cash it pays for itself only after 999999y 11mo: $999,999,999,999,999 worse off within the horizon.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 19y 11mo, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — but a caveat below contradicts the throughput this rests on. Until that is resolved, treat the payback as the earliest it could possibly arrive, not a forecast: at one token per step the rate is overstated, so is everything computed from it, and the real payback is longer than the figure above. In cash it pays for itself only after 999999y: $999,999,999,999,999 worse off within the horizon.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 19y 11mo, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — but a caveat below contradicts the throughput this rests on. Until that is resolved, treat the payback as the earliest it could possibly arrive, not a forecast: at one token per step the rate is overstated, so is everything computed from it, and the real payback is longer than the figure above. In cash it never pays for itself: $999,999,999,999,999 worse off over 19.99 years.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 19y, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — but a caveat below contradicts the throughput this rests on. Until that is resolved, treat the payback as the earliest it could possibly arrive, not a forecast: at one token per step the rate is overstated, so is everything computed from it, and the real payback is longer than the figure above. In cash it pays for itself after 19y 11mo and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 19y, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — but a caveat below contradicts the throughput this rests on. Until that is resolved, treat the payback as the earliest it could possibly arrive, not a forecast: at one token per step the rate is overstated, so is everything computed from it, and the real payback is longer than the figure above. In cash it pays for itself after 59d and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 19y, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — but a caveat below contradicts the throughput this rests on. Until that is resolved, treat the payback as the earliest it could possibly arrive, not a forecast: at one token per step the rate is overstated, so is everything computed from it, and the real payback is longer than the figure above. In cash it pays for itself after 23h 59m and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 19y, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — but a caveat below contradicts the throughput this rests on. Until that is resolved, treat the payback as the earliest it could possibly arrive, not a forecast: at one token per step the rate is overstated, so is everything computed from it, and the real payback is longer than the figure above. In cash it pays for itself only after 999999y 11mo: $999,999,999,999,999 worse off within the horizon.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 19y, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — but a caveat below contradicts the throughput this rests on. Until that is resolved, treat the payback as the earliest it could possibly arrive, not a forecast: at one token per step the rate is overstated, so is everything computed from it, and the real payback is longer than the figure above. In cash it pays for itself only after 999999y: $999,999,999,999,999 worse off within the horizon.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 19y, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — but a caveat below contradicts the throughput this rests on. Until that is resolved, treat the payback as the earliest it could possibly arrive, not a forecast: at one token per step the rate is overstated, so is everything computed from it, and the real payback is longer than the figure above. In cash it never pays for itself: $999,999,999,999,999 worse off over 19.99 years.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 59d, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — but a caveat below contradicts the throughput this rests on. Until that is resolved, treat the payback as the earliest it could possibly arrive, not a forecast: at one token per step the rate is overstated, so is everything computed from it, and the real payback is longer than the figure above. In cash it pays for itself after 19y 11mo and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 59d, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — but a caveat below contradicts the throughput this rests on. Until that is resolved, treat the payback as the earliest it could possibly arrive, not a forecast: at one token per step the rate is overstated, so is everything computed from it, and the real payback is longer than the figure above. In cash it pays for itself after 59d and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 59d, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — but a caveat below contradicts the throughput this rests on. Until that is resolved, treat the payback as the earliest it could possibly arrive, not a forecast: at one token per step the rate is overstated, so is everything computed from it, and the real payback is longer than the figure above. In cash it pays for itself after 23h 59m and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 59d, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — but a caveat below contradicts the throughput this rests on. Until that is resolved, treat the payback as the earliest it could possibly arrive, not a forecast: at one token per step the rate is overstated, so is everything computed from it, and the real payback is longer than the figure above. In cash it pays for itself only after 999999y 11mo: $999,999,999,999,999 worse off within the horizon.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 59d, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — but a caveat below contradicts the throughput this rests on. Until that is resolved, treat the payback as the earliest it could possibly arrive, not a forecast: at one token per step the rate is overstated, so is everything computed from it, and the real payback is longer than the figure above. In cash it pays for itself only after 999999y: $999,999,999,999,999 worse off within the horizon.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 59d, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — but a caveat below contradicts the throughput this rests on. Until that is resolved, treat the payback as the earliest it could possibly arrive, not a forecast: at one token per step the rate is overstated, so is everything computed from it, and the real payback is longer than the figure above. In cash it never pays for itself: $999,999,999,999,999 worse off over 19.99 years.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 23h 59m, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — but a caveat below contradicts the throughput this rests on. Until that is resolved, treat the payback as the earliest it could possibly arrive, not a forecast: at one token per step the rate is overstated, so is everything computed from it, and the real payback is longer than the figure above. In cash it pays for itself after 19y 11mo and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 23h 59m, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — but a caveat below contradicts the throughput this rests on. Until that is resolved, treat the payback as the earliest it could possibly arrive, not a forecast: at one token per step the rate is overstated, so is everything computed from it, and the real payback is longer than the figure above. In cash it pays for itself after 59d and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 23h 59m, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — but a caveat below contradicts the throughput this rests on. Until that is resolved, treat the payback as the earliest it could possibly arrive, not a forecast: at one token per step the rate is overstated, so is everything computed from it, and the real payback is longer than the figure above. In cash it pays for itself after 23h 59m and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 23h 59m, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — but a caveat below contradicts the throughput this rests on. Until that is resolved, treat the payback as the earliest it could possibly arrive, not a forecast: at one token per step the rate is overstated, so is everything computed from it, and the real payback is longer than the figure above. In cash it pays for itself only after 999999y 11mo: $999,999,999,999,999 worse off within the horizon.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 23h 59m, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — but a caveat below contradicts the throughput this rests on. Until that is resolved, treat the payback as the earliest it could possibly arrive, not a forecast: at one token per step the rate is overstated, so is everything computed from it, and the real payback is longer than the figure above. In cash it pays for itself only after 999999y: $999,999,999,999,999 worse off within the horizon.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 23h 59m, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — but a caveat below contradicts the throughput this rests on. Until that is resolved, treat the payback as the earliest it could possibly arrive, not a forecast: at one token per step the rate is overstated, so is everything computed from it, and the real payback is longer than the figure above. In cash it never pays for itself: $999,999,999,999,999 worse off over 19.99 years.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 23h, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — but a caveat below contradicts the throughput this rests on. Until that is resolved, treat the payback as the earliest it could possibly arrive, not a forecast: at one token per step the rate is overstated, so is everything computed from it, and the real payback is longer than the figure above. In cash it pays for itself after 19y 11mo and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 23h, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — but a caveat below contradicts the throughput this rests on. Until that is resolved, treat the payback as the earliest it could possibly arrive, not a forecast: at one token per step the rate is overstated, so is everything computed from it, and the real payback is longer than the figure above. In cash it pays for itself after 59d and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 23h, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — but a caveat below contradicts the throughput this rests on. Until that is resolved, treat the payback as the earliest it could possibly arrive, not a forecast: at one token per step the rate is overstated, so is everything computed from it, and the real payback is longer than the figure above. In cash it pays for itself after 23h 59m and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 23h, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — but a caveat below contradicts the throughput this rests on. Until that is resolved, treat the payback as the earliest it could possibly arrive, not a forecast: at one token per step the rate is overstated, so is everything computed from it, and the real payback is longer than the figure above. In cash it pays for itself only after 999999y 11mo: $999,999,999,999,999 worse off within the horizon.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 23h, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — but a caveat below contradicts the throughput this rests on. Until that is resolved, treat the payback as the earliest it could possibly arrive, not a forecast: at one token per step the rate is overstated, so is everything computed from it, and the real payback is longer than the figure above. In cash it pays for itself only after 999999y: $999,999,999,999,999 worse off within the horizon.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 23h, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — but a caveat below contradicts the throughput this rests on. Until that is resolved, treat the payback as the earliest it could possibly arrive, not a forecast: at one token per step the rate is overstated, so is everything computed from it, and the real payback is longer than the figure above. In cash it never pays for itself: $999,999,999,999,999 worse off over 19.99 years.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 19y 11mo, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which a caveat below finds beyond the checks' estimate without ruling it out. In cash it pays for itself after 19y 11mo and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 19y 11mo, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which a caveat below finds beyond the checks' estimate without ruling it out. In cash it pays for itself after 59d and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 19y 11mo, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which a caveat below finds beyond the checks' estimate without ruling it out. In cash it pays for itself after 23h 59m and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 19y 11mo, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which a caveat below finds beyond the checks' estimate without ruling it out. In cash it pays for itself only after 999999y 11mo: $999,999,999,999,999 worse off within the horizon.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 19y 11mo, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which a caveat below finds beyond the checks' estimate without ruling it out. In cash it pays for itself only after 999999y: $999,999,999,999,999 worse off within the horizon.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 19y 11mo, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which a caveat below finds beyond the checks' estimate without ruling it out. In cash it never pays for itself: $999,999,999,999,999 worse off over 19.99 years.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 19y, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which a caveat below finds beyond the checks' estimate without ruling it out. In cash it pays for itself after 19y 11mo and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 19y, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which a caveat below finds beyond the checks' estimate without ruling it out. In cash it pays for itself after 59d and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 19y, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which a caveat below finds beyond the checks' estimate without ruling it out. In cash it pays for itself after 23h 59m and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 19y, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which a caveat below finds beyond the checks' estimate without ruling it out. In cash it pays for itself only after 999999y 11mo: $999,999,999,999,999 worse off within the horizon.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 19y, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which a caveat below finds beyond the checks' estimate without ruling it out. In cash it pays for itself only after 999999y: $999,999,999,999,999 worse off within the horizon.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 19y, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which a caveat below finds beyond the checks' estimate without ruling it out. In cash it never pays for itself: $999,999,999,999,999 worse off over 19.99 years.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 59d, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which a caveat below finds beyond the checks' estimate without ruling it out. In cash it pays for itself after 19y 11mo and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 59d, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which a caveat below finds beyond the checks' estimate without ruling it out. In cash it pays for itself after 59d and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 59d, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which a caveat below finds beyond the checks' estimate without ruling it out. In cash it pays for itself after 23h 59m and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 59d, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which a caveat below finds beyond the checks' estimate without ruling it out. In cash it pays for itself only after 999999y 11mo: $999,999,999,999,999 worse off within the horizon.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 59d, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which a caveat below finds beyond the checks' estimate without ruling it out. In cash it pays for itself only after 999999y: $999,999,999,999,999 worse off within the horizon.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 59d, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which a caveat below finds beyond the checks' estimate without ruling it out. In cash it never pays for itself: $999,999,999,999,999 worse off over 19.99 years.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 23h 59m, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which a caveat below finds beyond the checks' estimate without ruling it out. In cash it pays for itself after 19y 11mo and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 23h 59m, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which a caveat below finds beyond the checks' estimate without ruling it out. In cash it pays for itself after 59d and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 23h 59m, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which a caveat below finds beyond the checks' estimate without ruling it out. In cash it pays for itself after 23h 59m and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 23h 59m, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which a caveat below finds beyond the checks' estimate without ruling it out. In cash it pays for itself only after 999999y 11mo: $999,999,999,999,999 worse off within the horizon.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 23h 59m, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which a caveat below finds beyond the checks' estimate without ruling it out. In cash it pays for itself only after 999999y: $999,999,999,999,999 worse off within the horizon.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 23h 59m, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which a caveat below finds beyond the checks' estimate without ruling it out. In cash it never pays for itself: $999,999,999,999,999 worse off over 19.99 years.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 23h, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which a caveat below finds beyond the checks' estimate without ruling it out. In cash it pays for itself after 19y 11mo and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 23h, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which a caveat below finds beyond the checks' estimate without ruling it out. In cash it pays for itself after 59d and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 23h, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which a caveat below finds beyond the checks' estimate without ruling it out. In cash it pays for itself after 23h 59m and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 23h, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which a caveat below finds beyond the checks' estimate without ruling it out. In cash it pays for itself only after 999999y 11mo: $999,999,999,999,999 worse off within the horizon.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 23h, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which a caveat below finds beyond the checks' estimate without ruling it out. In cash it pays for itself only after 999999y: $999,999,999,999,999 worse off within the horizon.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 23h, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which a caveat below finds beyond the checks' estimate without ruling it out. In cash it never pays for itself: $999,999,999,999,999 worse off over 19.99 years.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 19y 11mo, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is estimated evidence at medium confidence. In cash it pays for itself after 19y 11mo and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 19y 11mo, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is estimated evidence at medium confidence. In cash it pays for itself after 59d and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 19y 11mo, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is estimated evidence at medium confidence. In cash it pays for itself after 23h 59m and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 19y 11mo, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is estimated evidence at medium confidence. In cash it pays for itself only after 999999y 11mo: $999,999,999,999,999 worse off within the horizon.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 19y 11mo, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is estimated evidence at medium confidence. In cash it pays for itself only after 999999y: $999,999,999,999,999 worse off within the horizon.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 19y 11mo, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is estimated evidence at medium confidence. In cash it never pays for itself: $999,999,999,999,999 worse off over 19.99 years.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 19y, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is estimated evidence at medium confidence. In cash it pays for itself after 19y 11mo and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 19y, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is estimated evidence at medium confidence. In cash it pays for itself after 59d and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 19y, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is estimated evidence at medium confidence. In cash it pays for itself after 23h 59m and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 19y, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is estimated evidence at medium confidence. In cash it pays for itself only after 999999y 11mo: $999,999,999,999,999 worse off within the horizon.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 19y, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is estimated evidence at medium confidence. In cash it pays for itself only after 999999y: $999,999,999,999,999 worse off within the horizon.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 19y, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is estimated evidence at medium confidence. In cash it never pays for itself: $999,999,999,999,999 worse off over 19.99 years.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 59d, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is estimated evidence at medium confidence. In cash it pays for itself after 19y 11mo and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 59d, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is estimated evidence at medium confidence. In cash it pays for itself after 59d and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 59d, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is estimated evidence at medium confidence. In cash it pays for itself after 23h 59m and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 59d, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is estimated evidence at medium confidence. In cash it pays for itself only after 999999y 11mo: $999,999,999,999,999 worse off within the horizon.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 59d, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is estimated evidence at medium confidence. In cash it pays for itself only after 999999y: $999,999,999,999,999 worse off within the horizon.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 59d, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is estimated evidence at medium confidence. In cash it never pays for itself: $999,999,999,999,999 worse off over 19.99 years.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 23h 59m, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is estimated evidence at medium confidence. In cash it pays for itself after 19y 11mo and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 23h 59m, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is estimated evidence at medium confidence. In cash it pays for itself after 59d and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 23h 59m, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is estimated evidence at medium confidence. In cash it pays for itself after 23h 59m and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 23h 59m, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is estimated evidence at medium confidence. In cash it pays for itself only after 999999y 11mo: $999,999,999,999,999 worse off within the horizon.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 23h 59m, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is estimated evidence at medium confidence. In cash it pays for itself only after 999999y: $999,999,999,999,999 worse off within the horizon.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 23h 59m, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is estimated evidence at medium confidence. In cash it never pays for itself: $999,999,999,999,999 worse off over 19.99 years.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 23h, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is estimated evidence at medium confidence. In cash it pays for itself after 19y 11mo and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 23h, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is estimated evidence at medium confidence. In cash it pays for itself after 59d and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 23h, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is estimated evidence at medium confidence. In cash it pays for itself after 23h 59m and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 23h, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is estimated evidence at medium confidence. In cash it pays for itself only after 999999y 11mo: $999,999,999,999,999 worse off within the horizon.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 23h, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is estimated evidence at medium confidence. In cash it pays for itself only after 999999y: $999,999,999,999,999 worse off within the horizon.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 23h, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is estimated evidence at medium confidence. In cash it never pays for itself: $999,999,999,999,999 worse off over 19.99 years.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 19y 11mo, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is estimated evidence at low confidence. In cash it pays for itself after 19y 11mo and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 19y 11mo, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is estimated evidence at low confidence. In cash it pays for itself after 59d and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 19y 11mo, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is estimated evidence at low confidence. In cash it pays for itself after 23h 59m and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 19y 11mo, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is estimated evidence at low confidence. In cash it pays for itself only after 999999y 11mo: $999,999,999,999,999 worse off within the horizon.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 19y 11mo, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is estimated evidence at low confidence. In cash it pays for itself only after 999999y: $999,999,999,999,999 worse off within the horizon.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 19y 11mo, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is estimated evidence at low confidence. In cash it never pays for itself: $999,999,999,999,999 worse off over 19.99 years.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 19y, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is estimated evidence at low confidence. In cash it pays for itself after 19y 11mo and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 19y, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is estimated evidence at low confidence. In cash it pays for itself after 59d and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 19y, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is estimated evidence at low confidence. In cash it pays for itself after 23h 59m and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 19y, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is estimated evidence at low confidence. In cash it pays for itself only after 999999y 11mo: $999,999,999,999,999 worse off within the horizon.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 19y, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is estimated evidence at low confidence. In cash it pays for itself only after 999999y: $999,999,999,999,999 worse off within the horizon.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 19y, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is estimated evidence at low confidence. In cash it never pays for itself: $999,999,999,999,999 worse off over 19.99 years.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 59d, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is estimated evidence at low confidence. In cash it pays for itself after 19y 11mo and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 59d, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is estimated evidence at low confidence. In cash it pays for itself after 59d and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 59d, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is estimated evidence at low confidence. In cash it pays for itself after 23h 59m and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 59d, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is estimated evidence at low confidence. In cash it pays for itself only after 999999y 11mo: $999,999,999,999,999 worse off within the horizon.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 59d, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is estimated evidence at low confidence. In cash it pays for itself only after 999999y: $999,999,999,999,999 worse off within the horizon.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 59d, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is estimated evidence at low confidence. In cash it never pays for itself: $999,999,999,999,999 worse off over 19.99 years.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 23h 59m, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is estimated evidence at low confidence. In cash it pays for itself after 19y 11mo and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 23h 59m, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is estimated evidence at low confidence. In cash it pays for itself after 59d and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 23h 59m, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is estimated evidence at low confidence. In cash it pays for itself after 23h 59m and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 23h 59m, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is estimated evidence at low confidence. In cash it pays for itself only after 999999y 11mo: $999,999,999,999,999 worse off within the horizon.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 23h 59m, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is estimated evidence at low confidence. In cash it pays for itself only after 999999y: $999,999,999,999,999 worse off within the horizon.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 23h 59m, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is estimated evidence at low confidence. In cash it never pays for itself: $999,999,999,999,999 worse off over 19.99 years.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 23h, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is estimated evidence at low confidence. In cash it pays for itself after 19y 11mo and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 23h, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is estimated evidence at low confidence. In cash it pays for itself after 59d and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 23h, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is estimated evidence at low confidence. In cash it pays for itself after 23h 59m and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 23h, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is estimated evidence at low confidence. In cash it pays for itself only after 999999y 11mo: $999,999,999,999,999 worse off within the horizon.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 23h, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is estimated evidence at low confidence. In cash it pays for itself only after 999999y: $999,999,999,999,999 worse off within the horizon.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 23h, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is estimated evidence at low confidence. In cash it never pays for itself: $999,999,999,999,999 worse off over 19.99 years.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 19y 11mo, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is assumption evidence at low confidence. In cash it pays for itself after 19y 11mo and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 19y 11mo, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is assumption evidence at low confidence. In cash it pays for itself after 59d and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 19y 11mo, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is assumption evidence at low confidence. In cash it pays for itself after 23h 59m and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 19y 11mo, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is assumption evidence at low confidence. In cash it pays for itself only after 999999y 11mo: $999,999,999,999,999 worse off within the horizon.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 19y 11mo, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is assumption evidence at low confidence. In cash it pays for itself only after 999999y: $999,999,999,999,999 worse off within the horizon.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 19y 11mo, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is assumption evidence at low confidence. In cash it never pays for itself: $999,999,999,999,999 worse off over 19.99 years.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 19y, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is assumption evidence at low confidence. In cash it pays for itself after 19y 11mo and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 19y, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is assumption evidence at low confidence. In cash it pays for itself after 59d and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 19y, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is assumption evidence at low confidence. In cash it pays for itself after 23h 59m and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 19y, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is assumption evidence at low confidence. In cash it pays for itself only after 999999y 11mo: $999,999,999,999,999 worse off within the horizon.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 19y, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is assumption evidence at low confidence. In cash it pays for itself only after 999999y: $999,999,999,999,999 worse off within the horizon.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 19y, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is assumption evidence at low confidence. In cash it never pays for itself: $999,999,999,999,999 worse off over 19.99 years.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 59d, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is assumption evidence at low confidence. In cash it pays for itself after 19y 11mo and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 59d, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is assumption evidence at low confidence. In cash it pays for itself after 59d and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 59d, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is assumption evidence at low confidence. In cash it pays for itself after 23h 59m and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 59d, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is assumption evidence at low confidence. In cash it pays for itself only after 999999y 11mo: $999,999,999,999,999 worse off within the horizon.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 59d, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is assumption evidence at low confidence. In cash it pays for itself only after 999999y: $999,999,999,999,999 worse off within the horizon.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 59d, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is assumption evidence at low confidence. In cash it never pays for itself: $999,999,999,999,999 worse off over 19.99 years.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 23h 59m, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is assumption evidence at low confidence. In cash it pays for itself after 19y 11mo and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 23h 59m, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is assumption evidence at low confidence. In cash it pays for itself after 59d and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 23h 59m, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is assumption evidence at low confidence. In cash it pays for itself after 23h 59m and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 23h 59m, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is assumption evidence at low confidence. In cash it pays for itself only after 999999y 11mo: $999,999,999,999,999 worse off within the horizon.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 23h 59m, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is assumption evidence at low confidence. In cash it pays for itself only after 999999y: $999,999,999,999,999 worse off within the horizon.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 23h 59m, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is assumption evidence at low confidence. In cash it never pays for itself: $999,999,999,999,999 worse off over 19.99 years.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 23h, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is assumption evidence at low confidence. In cash it pays for itself after 19y 11mo and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 23h, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is assumption evidence at low confidence. In cash it pays for itself after 59d and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 23h, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is assumption evidence at low confidence. In cash it pays for itself after 23h 59m and is $999,999,999,999,999 ahead.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 23h, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is assumption evidence at low confidence. In cash it pays for itself only after 999999y 11mo: $999,999,999,999,999 worse off within the horizon.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 23h, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is assumption evidence at low confidence. In cash it pays for itself only after 999999y: $999,999,999,999,999 worse off within the horizon.At 100% utilization and 2.00× quality parity DGX B300 costs less than Anthropic Claude Opus 5.5 after 23h, and is $999,999,999,999,999 ahead by the end of the 19.99-year horizon — if your deployment reaches the modeled throughput, which is assumption evidence at low confidence. In cash it never pays for itself: $999,999,999,999,999 worse off over 19.99 years.
Read this result with 3 caveatsNo caveats left on this resultRead this result with one caveatRead this result with 2 caveatsRead this result with 3 caveats
This request's 20,000 input and 5,000 output tokens are not in the ratio the "base" profile's rates were split at, 8,192 to 1,024.
Request rate is the lower of decode ÷ output tokens and prefill ÷ input tokens, and the two limits meet only at the ratio the rates were split at. Here decode runs out first: 1,240 tokens/s ÷ 5,000 output tokens admits 893 requests an hour at full utilization, where prefill's 9,920 ÷ 20,000 admits 1,786, so 50% of the prefill capacity goes unused. A deployment built for this request would move GPU time from prefill to decode, and at the runs' cost per token it would serve between those two figures; the result is costed at the lower one, so on ratio alone it understates. At 25,000 tokens the request is also longer than the runs' 9,216, and every token of a longer request costs more in both phases, which the calculator does not model and which overstates.
Rates measured at this request's shape can replace the profile's under Throughput evidence.
Aggregate memory fits, but the topology is below official guidance.
Moonshot AI recommends at least 64 accelerators for Kimi K3; this scenario uses 8 × B300. A capacity check is not a statement that the deployment works or performs.
This scenario credits Kimi K3 with more value than Artificial Analysis Intelligence Index v4.3.2 supports.
Artificial Analysis Intelligence Index v4.3.2 puts Kimi K3 at 44 and Claude Opus 5.5 at 58 on the same scale, a ratio of 0.76. Quality parity is set to 1.00×, so the comparison treats one unit of Kimi K3 output as worth more than that scale suggests. Cost per token is only meaningful between models that can do the same job: a composite index is a blunt instrument, but a gap is worth pricing deliberately rather than by default.
Set quality parity to what the two models are worth for your workload — the index is a starting point, not an answer.
Cumulative cost curve
Self-hosted
APIAPI value at 1.00×
API bill
self-hosted API equivalent API value at 1.00× API bill
showing the whole horizon · drag across the chart to zoomshowing month 0 to 36 of 36
At month 36, self-hosted cost is $630,645 and API equivalent is $1,799,126. Break-even lands after 10mo, 26% of the way into the modeled horizon.
What the same workload costs on every current tier
Cost per request at 20,000 input and 5,000 output tokens, with the current cache ratios and routing. Token counts are not comparable
across tokenizers, so treat this as a spread, not a ranking of value. A tier marked estimated is priced partly from
rates derived here from other published figures rather than read off its provider's
page, and its note in the catalog says which. Google rows include no cache storage.Google rows include no cache storage.Google rows include 59.99 minutes of cache storage per request where a rate is recorded.
Solved thresholds, not sampled points: the value of each assumption at which break-even stops landing inside the 3-year horizon.Solved thresholds, not sampled points: the value of each assumption at which break-even, at 2.00× quality parity, stops landing inside the 19.75-year horizon.
Utilization must stay above
21%No value worksHolds throughout100%
assumed 60%assumed 100%
Achievable throughput must stay above
35%No value worksHolds throughout100%
assumed 100%assumed 100%
Quality parity must stay above
0.35×No value worksHolds throughout2.00×
assumed 1.00×assumed 2.00×
What moves the answer
Each driver varied on its own, with every other assumption held constant. Break-even and API value count the API side at quality parity, the scenario's or the row's own. In cash counts the API's own bill, $5.11 per million tokens in every row, and is left blank in rows at 1.00×, where it would repeat Break-even.
Driver
Break-even
In cash
Self-host $/M
API value $/M
Utilizationassumed 60%
25%
2y 4mo
2y 4mo
$4.30
$5.11
50%
12mo
12mo
$2.15
$5.11
75%
7mo
7mo
$1.43
$5.11
100%
5mo
5mo
$1.08
$5.11
Achievable throughputassumed 100%
50%
1y 10mo
1y 10mo
$3.58
$5.11
70%
1y 2mo
1y 2mo
$2.56
$5.11
85%
11mo
11mo
$2.11
$5.11
100%
10mo
10mo
$1.79
$5.11
Quality parityassumed 1.00×
0.75×
1y 1mo
10mo
$1.79
$3.83
1.00×
10mo
10mo
$1.79
$5.11
1.25×
7mo
10mo
$1.79
$6.39
Where the memory goes
Aggregate across 8 accelerators, holding 48 sequences of 25,000 tokens each.
Weights
1,561 GB
Runtime overhead
312 GB
KV cache
55 GB
Free
376 GB
Sequences that fit
377
Context that fits
308,489
Sequences and context that fit are counted with one recurrent state per sequence, where vLLM keeps two or three per request by default and SGLang admits one request per five state slots, so on either default they are the most that could fit.Sequences and context that fit are counted with one recurrent state per sequence, where vLLM keeps two or three per request by default and SGLang admits one request per five state slots, so on either default they are the most that could fit. The cache figures above also assume this model's heads split evenly across all 72 accelerators, which neither vLLM nor SGLang loads.Sequences and context that fit are counted with every cache this model's profile records, and with no memory a runtime sets aside beyond the runtime overhead above.Sequences and context that fit are counted with every cache this model's profile records, and with no memory a runtime sets aside beyond the runtime overhead above. The cache figures above also assume this model's heads split evenly across all 72 accelerators, which neither vLLM nor SGLang loads.Sequences and context that fit are counted without any storage the indexer's candidate pool needs, so they are the most that could fit.Sequences and context that fit are counted without any storage the indexer's candidate pool needs, so they are the most that could fit. The cache figures above also assume this model's heads split evenly across all 72 accelerators, which neither vLLM nor SGLang loads.This model publishes no cache geometry, so no sequence or context count is given, and any cache would have to fit in the free memory above.Sequences and context that fit are counted with one recurrent state per sequence, where vLLM keeps two or three per request by default and SGLang admits one request per five state slots, so on either default they are the most that could fit.
Is the cited rate reachable here
A check on the cited rate, not a forecast for your deployment: 1,000,000,000 tokens/s against 1,670 TB/s of published bandwidth and a 4,892 GB checkpoint.A check on the cited rate, not a forecast for your deployment: 1,240 tokens/s against 64 TB/s of published bandwidth and a 1,561 GB checkpoint.
Requests in flight the rate needs
2
The fewest it could need
2
Requests this system can hold
377
Cache reads alone
1.4 TB/s
Memory bandwidth available
64 TB/s
A decode step reads the weights its requests need, then some of every in-flight
request's cache. The weight read is shared by every request in the step, so a high
token rate comes from spreading it across many requests at once, or from emitting
several tokens a step. The first two rows are that arithmetic under two sets of
assumptions about how much a step reads; neither describes a particular serving
stack.
This model publishes no cache geometry, so how many of those requests fit is not modelled. See how this is derived.Even the least a step can read of each request's cache exceeds the memory system at this request size, so at one token per step no batch size reaches the rate. See how this is derived.Even reading only the active weights, a step exceeds the memory system at every batch size tried, so at one token per step the rate is out of reach. See how this is derived.Reading every weight, a step exceeds the memory system at every batch size tried, so at one token per step the rate is out of reach. See how this is derived.This system cannot hold even the fewest requests the rate could need, so at one token per step the rate and this request size cannot both hold. See how this is derived.This system holds the requests the estimate needs, so the rate and this request size are consistent. Requests this system can hold are counted with one recurrent state each, where vLLM and SGLang both keep more by default, so the system may hold fewer. See how this is derived.The estimate does not fit this system and the fewest requests the rate could need do, so the rate rests on a step reading less than the estimate assumes, or on several tokens per step. Requests this system can hold are counted with one recurrent state each, where vLLM and SGLang both keep more by default, so the system may hold fewer. See how this is derived.Even the least a step can read of each request's cache exceeds the memory system at this request size, so at one token per step no batch size reaches the rate. Every request count and cache read above assumes this model's heads split evenly across all 72 accelerators, which neither vLLM nor SGLang loads. See how this is derived.Even reading only the active weights, a step exceeds the memory system at every batch size tried, so at one token per step the rate is out of reach. Every request count and cache read above assumes this model's heads split evenly across all 72 accelerators, which neither vLLM nor SGLang loads. See how this is derived.Reading every weight, a step exceeds the memory system at every batch size tried, so at one token per step the rate is out of reach. Every request count and cache read above assumes this model's heads split evenly across all 72 accelerators, which neither vLLM nor SGLang loads. See how this is derived.This system cannot hold even the fewest requests the rate could need, so at one token per step the rate and this request size cannot both hold. Every request count and cache read above assumes this model's heads split evenly across all 72 accelerators, which neither vLLM nor SGLang loads. See how this is derived.This system holds the requests the estimate needs, so the rate and this request size are consistent. Requests this system can hold are counted with one recurrent state each, where vLLM and SGLang both keep more by default, so the system may hold fewer. Every request count and cache read above assumes this model's heads split evenly across all 72 accelerators, which neither vLLM nor SGLang loads. See how this is derived.The estimate does not fit this system and the fewest requests the rate could need do, so the rate rests on a step reading less than the estimate assumes, or on several tokens per step. Requests this system can hold are counted with one recurrent state each, where vLLM and SGLang both keep more by default, so the system may hold fewer. Every request count and cache read above assumes this model's heads split evenly across all 72 accelerators, which neither vLLM nor SGLang loads. See how this is derived.This system holds the requests the estimate needs, so the rate and this request size are consistent. See how this is derived.The estimate does not fit this system and the fewest requests the rate could need do, so the rate rests on a step reading less than the estimate assumes, or on several tokens per step. See how this is derived.This system holds the requests the estimate needs, so the rate and this request size are consistent. Every request count and cache read above assumes this model's heads split evenly across all 72 accelerators, which neither vLLM nor SGLang loads. See how this is derived.The estimate does not fit this system and the fewest requests the rate could need do, so the rate rests on a step reading less than the estimate assumes, or on several tokens per step. Every request count and cache read above assumes this model's heads split evenly across all 72 accelerators, which neither vLLM nor SGLang loads. See how this is derived.This system holds the requests the estimate needs, so the rate and this request size are consistent. Every count above leaves out any storage the indexer's candidate pool needs, so the system may hold fewer requests and the rate may need more of them in flight. See how this is derived.The estimate does not fit this system and the fewest requests the rate could need do, so the rate rests on a step reading less than the estimate assumes, or on several tokens per step. Every count above leaves out any storage the indexer's candidate pool needs, so the system may hold fewer requests and the rate may need more of them in flight. See how this is derived.This system holds the requests the estimate needs, so the rate and this request size are consistent. Every count above leaves out any storage the indexer's candidate pool needs, so the system may hold fewer requests and the rate may need more of them in flight. Every request count and cache read above assumes this model's heads split evenly across all 72 accelerators, which neither vLLM nor SGLang loads. See how this is derived.The estimate does not fit this system and the fewest requests the rate could need do, so the rate rests on a step reading less than the estimate assumes, or on several tokens per step. Every count above leaves out any storage the indexer's candidate pool needs, so the system may hold fewer requests and the rate may need more of them in flight. Every request count and cache read above assumes this model's heads split evenly across all 72 accelerators, which neither vLLM nor SGLang loads. See how this is derived.This system holds the requests the estimate needs, so the rate and this request size are consistent. Requests this system can hold are counted with one recurrent state each, where vLLM and SGLang both keep more by default, so the system may hold fewer. See how this is derived.
Hourly economics
At the selected workload and utilization.
API-equivalent value
$68.46$999,999,999.99-$999,999,999.99
Raw API spend
$68.46
Electricity
$2.26
Support
$5.75
Net savingsNet value
$60.44
Net cash
$60.44
02 · inspect the inputs
Catalog
Every value is versioned, dated, and tagged with the kind of evidence behind it.
26hardware systems
37open models
83API price tiers
42benchmark observations
302sources
Browse the full catalog with specifications, prices, evidence badges and source links —
researched to 2026-10-07.
Reproducible formulas, explicit exclusions, and a clear quality caveat.
What is calculated
Per-request input, cache-hit, cache-write, and output tokens are priced separately. Throughput
is constrained by both decode and prefill; facility electricity applies IT power × PUE. IT
power is the whole system at full load, charged for every hour of the horizon: where a
published figure counts the accelerators alone, the catalog adds a cited allowance for the
host, and each system's catalog card says which basis its figure is on. Support is an annual
ratio of all-in acquisition cost. Break-even exists only when the hourly API-equivalent value,
the API bill for the same tokens times quality parity, exceeds self-hosted variable cost.
Memory, including KV cache
Weights fitting is necessary, not sufficient. The memory check covers weights plus a runtime
overhead ratio and the KV cache, which grows with context length and with how many requests
are in flight — so a model can pass the capacity check and still be unable to hold more than a couple
of concurrent sequences. KV cost is summed over layer groups rather than assumed uniform, because
current flagships mix attention types: a sliding-window layer's cache stops growing at the window
and a linear-attention layer's state never grows at all. One consequence is worth stating plainly
— a compressed latent cache is a single key/value head, so ordinary tensor parallelism gives every
accelerator a full copy of it. The weights divide across the system and the cache does not. Models
with no published cache profile say so rather than guessing.
Every system's capacity is its vendor's own figure, and vendors count memory in binary
gigabytes: AMD's specification table labels it GiB, and NVIDIA's documentation shows the 80 GB
H100 reporting 85.5 billion bytes. Weights and caches are counted here in decimal gigabytes, a
billion bytes each. So every system holds about 7% more than the figure the check compares
against, and the check errs toward saying a model does not fit. It errs the same way for every
vendor, so no system gains on another.
Evidence, ranked
Values carry the kind of evidence behind them, strongest first. Measured means a run happened and its result was recorded: MLPerf's closed division, a third-party harness,
or logged traffic. It does not mean independent — that is a property of the source, and the Evidence
page names the publisher beside every claim and grades each source's independence in the registry
above them. One of the sources graded this way is this repository owner's own recorded agent traffic. Vendor-reported is a real measurement published by the party that built the
product, engine or build — usually on a stack it tuned, so it is an upper bound rather than a
promise. A quantizer's accuracy table for its own checkpoint is its own run, so it counts. Official claim is a published specification or list price: a design fact, not a run of a workload. Estimated is derived here from other published values, or is a third party's own estimate, such as a capability
score Artificial Analysis marks as one; assumption is a planning number this repository
chose. Where better public evidence exists, it replaces a vendor's own figure.
Achievable throughput
Every cost per token here scales inversely with the throughput you actually reach, and
published throughput figures are tuned figures. The common worry is that open-source serving
stacks lag a vendor's own; the best same-hardware evidence does not support it — in MLPerf
Inference v6.0 a tuned vLLM submission led its 8× B200 category outright, and retuning that
same engine moved its own result by about a fifth. What the data does show is a gap between a
tuned deployment and an untuned one. The achievable-throughput control models that gap
explicitly, and is kept separate from utilization, which is only how busy the hardware is
kept.
What the capability score is, and is not
Scores shown beside models are the Artificial Analysis Intelligence Index,
cited with a link to the model's own page there. They are that organisation's measurement, not
this repository's, or its estimate where it says so, which the badge marks with ≈. They carry
real limits. A composite collapses ten dissimilar evaluations — agentic tool use, terminal
coding, PhD-level science questions, long-context reasoning, hallucination rate — into one
number, so two models can tie on the headline and differ sharply at the thing you actually
need. The index is also a moving target: it has been revised at least fifteen times since
January 2024, four times in September 2026 alone, and a revision can change which benchmarks
count, how they are weighted, or even which model grades the open-ended answers, so figures
from different versions are not strictly comparable. And a score belongs to one specific
configuration — GPT-6 Sol scores 48 at its highest reasoning effort and 28 with reasoning off,
and where Artificial Analysis lists several, every score here is a model's highest reasoning
effort — while a self-hosted deployment can differ in quantization, reasoning effort and
context length from whatever was benchmarked.
Colour bands compare each score to the best in this catalog rather than to the index's nominal
ceiling of 100, because today's frontier sits in the low sixties. The score is here for one
reason: every headline on this page is a cost ratio, and a cost ratio between two models of
very different capability is a category error wearing the clothes of a finding.
Quality-parity caveat
The quality-parity multiplier scales the value of successful local output against the API
comparator. It does not claim the models are equivalent, and it never changes physical
throughput.
Exclusions and semantics
Financing, taxes, staffing, downtime, resale, opportunity cost, and production support beyond
the configured ratio are excluded. Tokenizers also differ per provider and even per model
generation — Anthropic documents that its newest generation produces roughly 30% more tokens
for the same text than Sonnet 4.6 and earlier — so equal token counts across API tiers do not
describe equal text volumes; affected tiers carry a note. Superseded catalog entries stay
available so older shared scenarios keep resolving, but are hidden until you ask for them. The
research cutoff is the catalog's last-updated boundary; changing the URL scenario changes
assumptions, not the evidence. The repository's docs/architecture.md and docs/PROJECT.md record the full scope.
Independent work this relies on
Two outside efforts do work this repository depends on and has not reproduced. Artificial Analysis produces every capability score shown beside a model or a tier;
each one links to the page it was read from, and what a composite index can and cannot tell you
is set out above.
Sebastian Raschka publishes per-release architecture notes and an LLM architecture gallery that read new releases down to their attention geometry. Both uses of it here are worth naming.
The first is corroboration: his read of GLM-5.3-Flash found the same 34 linear-attention layers
against 11 sparse-attention layers this catalog derived from the release's own configuration file,
from the same release and with no contact between the two.
The second is disagreement, which was more useful. His published cache-per-token figures match
this catalog exactly on Muse Glimmer 30B at 52 KiB and Qwen3.6 27B at 64 KiB, and differ on
Gemma 4 31B: 840 KiB against 880 KiB here. The entire 40 KiB gap is the ten global layers,
which his figure sizes at half what this catalog does. Two different readings of Gemma 4's
configuration land on that half — using the model-wide head dimension where the global layers
declare their own larger one, or treating its attention_k_eq_v flag as storing keys
and values in one tensor — and this catalog carried the second until 2026-08-21. Which of the two
is behind the published figure cannot be told from the figure, so the entry says so rather than
picking one. Both traps are now written out in the entry itself.