MiniMax M3 428B — B200 vs H200
Head-to-head AI inference benchmark comparison of B200 (NVIDIA Blackwell) and H200 (NVIDIA Hopper) on MiniMax M3 428B. 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 82 tok/s/user on MiniMax M3 428B: B200 hits 1683 tok/s/chip, H200 hits 666. Per-million costs land at $0.29 and $0.51 respectively. B200 is 76% cheaper per token; B200 delivers 153% more tok/s/chip.
B200 / H200 on MiniMax M3 428B at 152 tok/s/user: 1050 / 333 tok/s/chip, $0.46 / $1.00 per million tokens. B200 is 120% cheaper per token; B200 delivers 216% more tok/s/chip.
Toward the upper edge of the 11–294 tok/s/user interactivity band, at 223 tok/s/user on MiniMax M3 428B: B200 runs 445 tok/s/chip at $1.08/M tokens, H200 runs 166 at $2.10/M. B200 is 94% cheaper per token; B200 delivers 168% more tok/s/chip. (Numbers reflect the default 1k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | B200:1683.2H200:666.5 | B200:1050.5H200:332.6 | B200:444.6H200:166.1 |
| Cost ($/M tok) | B200:$0.288H200:$0.505 | B200:$0.457H200:$1.004 | B200:$1.081H200:$2.102 |
| tok/s/MW | B200:984302H200:486476 | B200:614321H200:242783 | B200:260001H200:121223 |
| Concurrency | B200:~62H200:~18 | B200:~16H200:~4 | B200:~4H200:~3 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.