Qwen 3.5 397B-A17B · Chip comparison
Qwen 3.5 397B-A17B — B200 vs GB300 NVL72
Head-to-head AI inference benchmark comparison of B200 (NVIDIA Blackwell) and GB300 NVL72 (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.
AgentX replays real coding-agent sessions rather than fixed-length prompts, so context grows turn over turn and most of each request is served from cache instead of being recomputed. That turns the comparison into a systems question: KV transfer between nodes, prefix-aware routing, and cache capacity all move the curve alongside raw chip throughput. Fixed-sequence workloads stay the clean baseline for kernel and silicon performance, so the two scenarios answer different questions about the same hardware. Learn more about AgentX →
Throughput at 141 tok/s/user on Qwen 3.5 397B-A17B: B200 hits 31,174 tok/s/chip, GB300 NVL72 hits 59,758. Per-million costs land at $0.015 and $0.011 respectively. GB300 NVL72 is 44% cheaper per token; GB300 NVL72 delivers 92% more tok/s/chip.
B200 / GB300 NVL72 on Qwen 3.5 397B-A17B at 214 tok/s/user: 19,213 / 33,238 tok/s/chip, $0.025 / $0.019 per million tokens. GB300 NVL72 is 30% cheaper per token; GB300 NVL72 delivers 73% more tok/s/chip.
Toward the upper edge of the 68–359 tok/s/user interactivity band, at 287 tok/s/user on Qwen 3.5 397B-A17B: B200 runs 8,540 tok/s/chip at $0.056/M tokens, GB300 NVL72 runs 24,883 at $0.026/M. GB300 NVL72 is 118% cheaper per token; GB300 NVL72 delivers 191% more tok/s/chip. (Numbers reflect the default agentic-traces · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
Interpolated from real benchmark data. Edit target interactivity values below to compare at different operating points.
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | B20031,173.9GB300 NVL7259,757.7 | B20019,213.0GB300 NVL7233,237.9 | B2008,539.9GB300 NVL7224,882.6 |
| Cost ($/M tok) | B200$0.015GB300 NVL72$0.011 | B200$0.025GB300 NVL72$0.019 | B200$0.056GB300 NVL72$0.026 |
| tok/s/MW | B20018,230,323GB300 NVL7228,187,615 | B20011,235,650GB300 NVL7215,678,249 | B2004,994,091GB300 NVL7211,737,088 |
| Concurrency | B200~23GB300 NVL72~392 | B200~29GB300 NVL72~159 | B200~11GB300 NVL72~113 |