MiniMax M2.5/M2.7 — B300 vs H200
Head-to-head AI inference benchmark comparison of B300 (NVIDIA Blackwell) and H200 (NVIDIA Hopper) on MiniMax M2.5/M2.7. Latency, throughput, and cost across LLM workloads. Use the chart controls below to switch sequences, precisions, and metrics — same interactions as the main inference chart.
Throughput at 52 tok/s/user on MiniMax M2.5/M2.7: B300 hits 4734 tok/s/GPU, H200 hits 1132. Per-million costs land at $0.14 and $0.35 respectively. B300 is 152% cheaper per token; B300 delivers 318% more tok/s/GPU.
B300 / H200 on MiniMax M2.5/M2.7 at 80 tok/s/user: 1504 / 469 tok/s/GPU, $0.44 / $0.83 per million tokens. B300 is 90% cheaper per token; B300 delivers 221% more tok/s/GPU.
Toward the upper edge of the 25–134 tok/s/user interactivity band, at 107 tok/s/user on MiniMax M2.5/M2.7: B300 runs 903 tok/s/GPU at $0.71/M tokens, H200 runs 310 at $1.27/M. B300 is 78% cheaper per token; B300 delivers 191% more tok/s/GPU. (Numbers reflect the default 1k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | B300:4734.2H200:1131.6 | B300:1503.7H200:468.6 | B300:902.7H200:309.7 |
| Cost ($/M tok) | B300:$0.138H200:$0.348 | B300:$0.437H200:$0.833 | B300:$0.711H200:$1.266 |
| tok/s/MW | B300:2181652H200:654108 | B300:692958H200:270858 | B300:415985H200:179025 |
| Concurrency | B300:~478H200:~64 | B300:~42H200:~19 | B300:~17H200:~6 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.