MiniMax M3 428B — B200 vs GB300 NVL72
Head-to-head AI inference benchmark comparison of B200 (NVIDIA Blackwell) and GB300 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.
B200 / GB300 NVL72 on MiniMax M3 428B at 42 tok/s/user: 3022 / 1607 tok/s/GPU, $0.18 / $0.42 per million tokens. B200 is 136% cheaper per token; B200 delivers 88% more tok/s/GPU.
Around the middle of the 9–142 tok/s/user interactivity band, at 75 tok/s/user on MiniMax M3 428B: B200 runs 1444 tok/s/GPU at $0.38/M tokens, GB300 NVL72 runs 466 at $1.55/M. B200 is 313% cheaper per token; B200 delivers 210% more tok/s/GPU.
Setting 109 tok/s/user as the target on MiniMax M3 428B, B200 produces 846 tok/s/GPU ($0.64 per million tokens) and GB300 NVL72 produces 185 ($3.79). B200 is 490% cheaper per token; B200 delivers 357% more tok/s/GPU. (Numbers reflect the default 1k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | B200:3021.5GB300 NVL72:1607.2 | B200:1443.8GB300 NVL72:465.7 | B200:845.8GB300 NVL72:185.3 |
| Cost ($/M tok) | B200:$0.179GB300 NVL72:$0.423 | B200:$0.376GB300 NVL72:$1.552 | B200:$0.642GB300 NVL72:$3.791 |
| tok/s/MW | B200:1392418GB300 NVL72:765340 | B200:665323GB300 NVL72:221750 | B200:389773GB300 NVL72:88227 |
| Concurrency | B200:~161GB300 NVL72:~225 | B200:~64GB300 NVL72:~22 | B200:~19GB300 NVL72:~5 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.