MiniMax M3 428B — B200 vs GB200 NVL72
Head-to-head AI inference benchmark comparison of B200 (NVIDIA Blackwell) and GB200 NVL72 (NVIDIA Blackwell) 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.
Near the low end of the 17–157 tok/s/user interactivity band, at 52 tok/s/user on MiniMax M3 428B: B200 runs 3004 tok/s/GPU at $0.18/M tokens, GB200 NVL72 runs 1598 at $0.38/M. B200 is 109% cheaper per token; B200 delivers 88% more tok/s/GPU.
Setting 87 tok/s/user as the target on MiniMax M3 428B, B200 produces 1607 tok/s/GPU ($0.35 per million tokens) and GB200 NVL72 produces 638 ($1.00). B200 is 187% cheaper per token; B200 delivers 152% more tok/s/GPU.
At 122 tok/s/user interactivity on MiniMax M3 428B, B200 delivers 1364 tok/s/GPU at $0.42 per million tokens; GB200 NVL72 delivers 211 tok/s/GPU at $2.86. B200 is 584% cheaper per token; B200 delivers 546% more tok/s/GPU at this point. (Numbers reflect the default 1k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | B200:3004.4GB200 NVL72:1598.2 | B200:1607.5GB200 NVL72:638.3 | B200:1363.5GB200 NVL72:210.9 |
| Cost ($/M tok) | B200:$0.183GB200 NVL72:$0.382 | B200:$0.347GB200 NVL72:$0.996 | B200:$0.419GB200 NVL72:$2.862 |
| tok/s/MW | B200:1384529GB200 NVL72:761024 | B200:740783GB200 NVL72:303973 | B200:628344GB200 NVL72:100447 |
| Concurrency | B200:~142GB200 NVL72:~180 | B200:~56GB200 NVL72:~74 | B200:~24GB200 NVL72:~18 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.