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 3226 tok/s/GPU for $0.17 per million tokens at 69 tok/s/user on Qwen 3.5 397B-A17B; H200 posts 485 tok/s/GPU for $0.81. B200 is 377% cheaper per token; B200 delivers 565% more tok/s/GPU.
Throughput at 107 tok/s/user on Qwen 3.5 397B-A17B: B200 hits 1555 tok/s/GPU, H200 hits 349. Per-million costs land at $0.35 and $1.11 respectively. B200 is 217% cheaper per token; B200 delivers 346% more tok/s/GPU.
B200 / H200 on Qwen 3.5 397B-A17B at 145 tok/s/user: 1001 / 282 tok/s/GPU, $0.54 / $1.39 per million tokens. B200 is 157% cheaper per token; B200 delivers 255% more tok/s/GPU. (Numbers reflect the default 1k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | B200:3225.9H200:485.3 | B200:1554.9H200:348.8 | B200:1001.4H200:282.0 |
| Cost ($/M tok) | B200:$0.169H200:$0.807 | B200:$0.349H200:$1.108 | B200:$0.540H200:$1.387 |
| tok/s/MW | B200:1486611H200:280535 | B200:716542H200:201595 | B200:461491H200:163002 |
| Concurrency | B200:~102H200:~28 | B200:~33H200:~13 | B200:~15H200:~8 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.