MiniMax M2.5/M2.7 — B300 vs H100
Head-to-head AI inference benchmark comparison of B300 (NVIDIA Blackwell) and H100 (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.
B300 / H100 on MiniMax M2.5/M2.7 at 60 tok/s/user: 3427 / 599 tok/s/chip, $0.18 / $0.53 per million tokens. B300 is 192% cheaper per token; B300 delivers 472% more tok/s/chip.
Around the middle of the 41–117 tok/s/user interactivity band, at 79 tok/s/user on MiniMax M2.5/M2.7: B300 runs 1534 tok/s/chip at $0.42/M tokens, H100 runs 353 at $0.92/M. B300 is 122% cheaper per token; B300 delivers 335% more tok/s/chip.
Setting 98 tok/s/user as the target on MiniMax M2.5/M2.7, B300 produces 1067 tok/s/chip ($0.57 per million tokens) and H100 produces 207 ($1.56). B300 is 174% cheaper per token; B300 delivers 415% more tok/s/chip. (Numbers reflect the default 1k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | B300:3426.9H100:599.4 | B300:1533.8H100:353.0 | B300:1066.6H100:207.1 |
| Cost ($/M tok) | B300:$0.183H100:$0.534 | B300:$0.415H100:$0.920 | B300:$0.570H100:$1.563 |
| tok/s/MW | B300:1803650H100:437502 | B300:807276H100:257637 | B300:561350H100:151188 |
| Concurrency | B300:~195H100:~41 | B300:~45H100:~18 | B300:~22H100:~9 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.