MiniMax M3 428B — B200 vs H100
Head-to-head AI inference benchmark comparison of B200 (NVIDIA Blackwell) and H100 (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.
B200 / H100 on MiniMax M3 428B at 61 tok/s/user: 792 / 724 tok/s/GPU, $0.69 / $0.50 per million tokens. H100 is 37% cheaper per token; B200 delivers 9% more tok/s/GPU.
Around the middle of the 15–198 tok/s/user interactivity band, at 107 tok/s/user on MiniMax M3 428B: B200 runs 331 tok/s/GPU at $1.63/M tokens, H100 runs 313 at $1.13/M. H100 is 44% cheaper per token; B200 delivers 6% more tok/s/GPU.
Setting 153 tok/s/user as the target on MiniMax M3 428B, B200 produces 125 tok/s/GPU ($4.32 per million tokens) and H100 produces 211 ($1.67). H100 is 159% cheaper per token; H100 delivers 68% more tok/s/GPU. (Numbers reflect the default 1k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | B200:791.9H100:724.2 | B200:331.0H100:313.3 | B200:125.4H100:210.5 |
| Cost ($/M tok) | B200:$0.687H100:$0.503 | B200:$1.627H100:$1.131 | B200:$4.319H100:$1.668 |
| tok/s/MW | B200:364911H100:418608 | B200:152538H100:181073 | B200:57797H100:121695 |
| Concurrency | B200:~29H100:~52 | B200:~8H100:~13 | B200:~4H100:~6 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.