Qwen 3.5 397B-A17B — B200 vs H200
Head-to-head AI inference benchmark comparison of B200 (NVIDIA Blackwell) and H200 (NVIDIA Hopper) on Qwen 3.5 397B-A17B. 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 posts 9041 tok/s/chip for $0.05 per million tokens at 60 tok/s/user on Qwen 3.5 397B-A17B; H200 posts 1688 tok/s/chip for $0.20. B200 is 277% cheaper per token; B200 delivers 435% more tok/s/chip.
Throughput at 103 tok/s/user on Qwen 3.5 397B-A17B: B200 hits 6719 tok/s/chip, H200 hits 1266. Per-million costs land at $0.07 and $0.27 respectively. B200 is 276% cheaper per token; B200 delivers 431% more tok/s/chip.
B200 / H200 on Qwen 3.5 397B-A17B at 145 tok/s/user: 5163 / 966 tok/s/chip, $0.09 / $0.35 per million tokens. B200 is 275% cheaper per token; B200 delivers 435% more tok/s/chip. (Numbers reflect the default 8k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | B200:9040.6H200:1688.3 | B200:6718.8H200:1265.7 | B200:5162.7H200:965.8 |
| Cost ($/M tok) | B200:$0.053H200:$0.201 | B200:$0.072H200:$0.269 | B200:$0.093H200:$0.349 |
| tok/s/MW | B200:5286919H200:1232355 | B200:3929107H200:923836 | B200:3019119H200:704957 |
| Concurrency | B200:~72H200:~25 | B200:~32H200:~11 | B200:~18H200:~6 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.