MiniMax M2.5/M2.7 — B200 vs B300
Head-to-head AI inference benchmark comparison of B200 (NVIDIA Blackwell) and B300 (NVIDIA Blackwell) 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 66 tok/s/user on MiniMax M2.5/M2.7: B200 hits 8554 tok/s/chip, B300 hits 7415. Per-million costs land at $0.06 and $0.09 respectively. B200 is 52% cheaper per token; B200 delivers 15% more tok/s/chip.
B200 / B300 on MiniMax M2.5/M2.7 at 110 tok/s/user: 2744 / 2602 tok/s/chip, $0.18 / $0.24 per million tokens. B200 is 38% cheaper per token; B200 delivers 5% more tok/s/chip.
Toward the upper edge of the 22–197 tok/s/user interactivity band, at 154 tok/s/user on MiniMax M2.5/M2.7: B200 runs 772 tok/s/chip at $0.63/M tokens, B300 runs 790 at $0.79/M. B200 is 26% cheaper per token; B300 delivers 2% more tok/s/chip. (Numbers reflect the default 1k/1k · fp4 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | B200:8553.8B300:7414.7 | B200:2744.2B300:2602.4 | B200:772.0B300:790.4 |
| Cost ($/M tok) | B200:$0.056B300:$0.085 | B200:$0.175B300:$0.241 | B200:$0.630B300:$0.793 |
| tok/s/MW | B200:5002251B300:3902450 | B200:1604791B300:1369681 | B200:451468B300:416006 |
| Concurrency | B200:~1000B300:~524 | B200:~128B300:~68 | B200:~12B300:~14 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.