Qwen 3.5 397B-A17B — B200 vs B300
Head-to-head AI inference benchmark comparison of B200 (NVIDIA Blackwell) and B300 (NVIDIA Blackwell) 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.
Throughput at 69 tok/s/user on Qwen 3.5 397B-A17B: B200 hits 8445 tok/s/chip, B300 hits 6029. Per-million costs land at $0.06 and $0.10 respectively. B200 is 83% cheaper per token; B200 delivers 40% more tok/s/chip.
B200 / B300 on Qwen 3.5 397B-A17B at 123 tok/s/user: 5903 / 3800 tok/s/chip, $0.08 / $0.16 per million tokens. B200 is 102% cheaper per token; B200 delivers 55% more tok/s/chip.
Toward the upper edge of the 16–230 tok/s/user interactivity band, at 177 tok/s/user on Qwen 3.5 397B-A17B: B200 runs 4296 tok/s/chip at $0.11/M tokens, B300 runs 2678 at $0.23/M. B200 is 111% cheaper per token; B200 delivers 60% more tok/s/chip. (Numbers reflect the default 8k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | B200:8445.3B300:6028.8 | B200:5903.5B300:3799.8 | B200:4296.4B300:2678.4 |
| Cost ($/M tok) | B200:$0.057B300:$0.104 | B200:$0.081B300:$0.165 | B200:$0.111B300:$0.235 |
| tok/s/MW | B200:4938783B300:3173049 | B200:3452324B300:1999883 | B200:2512489B300:1409673 |
| Concurrency | B200:~58B300:~41 | B200:~23B300:~15 | B200:~12B300:~7 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.