Calculator
MiMo-V2.5: self-host or API?
Self-hosting
Derived (an estimate)Estimate. The API is cheaper below 3,600 million tokens a day. Above that, self-hosting wins only if the configuration is busy at least 50% of the time, and only at our higher throughput estimates; at our lowest estimate, even constant use would not pay off.
At 50% utilization · Cheapest API offer: $0.149 per million tokens at Vercel AI Gateway (3:1 input:output mix)
Hardware rental only: the hourly price the provider publishes for the machine. It does not count the engineering to deploy and run the model, monitoring, redundancy, start-up time, or storage and network costs billed separately.
- Break-even volume
- 3,600 million tokens a day
- Throughput self-hosting needs
- 83,334 tokens per second
- At 50% utilization
- Throughput self-hosting needs
- 41,667 tokens per second
- If busy all the time
At least this many input plus output tokens per second, all GPUs together, while serving.
What published figures show
- Break-even utilization: Prudent Never (above 100%) · Median Never (above 100%) · Optimistic 49.2%
- Published figure used (opens in a new tab)
Rented configuration
8 x A100 SXM 80GB at Lambda
$535.68 a day · $22.32 an hour · Guaranteed · 640 GB of GPU memory · fp8
Cheapest configuration whose weights fit: 8 x RTX PRO 6000 96GB at CoreWeave, $480.00 a day
Checked Oct 31, 2026 Price page at the provider (opens in a new tab)
Calculation detail
Throughput scenarios
| Scenario | Tokens per second | Usable capacity (million tokens a day) | Break-even utilization | Self-hosted cost per million tokens | Against the cheapest API offer |
|---|---|---|---|---|---|
| Prudent | 14,553.9 | 628.73 | Never (above 100%) | $0.852 | Dearer |
| Median | 36,384.7 | 1,571.82 | Never (above 100%) | $0.341 | Dearer |
| Optimistic | 84,677 | 3,658.05 | 49.2% | $0.146 | Cheaper |
Cheapest of 9 current standard offers, promotions, unstable prices, off-peak prices without published hours and unverified prices far below those of the other providers excluded Checked Oct 31, 2026
Assumptions
- same architecture type, scaled by memory bandwidth and active weights (x1.53)
- Selected over the cheapest configuration (RTX PRO 6000 96GB) because its throughput is derived (an estimate); it costs 11.6% more per hour, within our 25% price tolerance.
- Hardware rental only: the hourly price the provider publishes for the machine. It does not count the engineering to deploy and run the model, monitoring, redundancy, start-up time, or storage and network costs billed separately.
- Configuration: we pick the cheapest one whose weights fit. If a configuration costing up to 25% more has a measured throughput, we pick it instead; failing that, one with a derived throughput; failing that, one with a published lower bound that settles the verdict; among equals, the cheaper. A configuration whose throughput is only a wide estimate is never preferred over the cheapest. This is our rule, not a technical requirement.
- Quantization fp8; guaranteed tier prices as published by the provider; utilization 50%.
- Published figures are generally peak throughput measured at high concurrency; an interactive service with latency targets sustains less.
- Throughput is aggregate (input and output tokens, 3:1). Published figures in generated tokens are converted with a prefill/generation speed ratio of 3 (prudent), 5 (median) and 8 (optimistic). We treat a published output figure as if the GPUs only generated tokens, ignoring that the benchmark also read prompts: this understates the throughput of the whole configuration, which is the cautious side for self-hosting.
- Break-even = daily GPU cost / API price per million tokens (3,600 million tokens a day). Self-hosting is cheaper only if the average utilization you can sustain exceeds break-even / raw capacity (the break-even utilization of each scenario), whatever the number of GPUs.