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.
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 99 tok/s/user on Qwen 3.5 397B-A17B: B200 hits 36255 tok/s/chip, B300 hits 48060. Per-million costs land at $0.01 and $0.01 respectively. B300 is 1% cheaper per token; B300 delivers 33% more tok/s/chip.
B200 / B300 on Qwen 3.5 397B-A17B at 184 tok/s/user: 21134 / 21778 tok/s/chip, $0.02 / $0.03 per million tokens. B200 is 27% cheaper per token; B300 delivers 3% more tok/s/chip.
Toward the upper edge of the 14–354 tok/s/user interactivity band, at 269 tok/s/user on Qwen 3.5 397B-A17B: B200 runs 12623 tok/s/chip at $0.04/M tokens, B300 runs 10109 at $0.06/M. B200 is 63% cheaper per token; B200 delivers 25% more tok/s/chip. (Numbers reflect the default agentic-traces · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | B200:36255.3B300:48060.5 | B200:21134.5B300:21778.2 | B200:12622.8B300:10109.1 |
| Cost ($/M tok) | B200:$0.013B300:$0.013 | B200:$0.023B300:$0.029 | B200:$0.038B300:$0.062 |
| tok/s/MW | B200:21201931B300:25294990 | B200:12359341B300:11462201 | B200:7381755B300:5320575 |
| Concurrency | B200:~28B300:~46 | B200:~22B300:~23 | B200:~16B300:~7 |
Agentic inference metrics from the AgentX scenario and fixed-sequence inference metrics across models, hardware configurations, and serving parameters.
TCO $/chip/hr:
Source: SemiAnalysis Market July 2026 Pricing Surveys & AI Cloud TCO Model
Note: Disaggregated inference configurations (e.g., MoRI SGLang, Dynamo TRTLLM) calculate cost per million tokens per decode chip or per prefill chip, rather than per total chip count. This makes direct token cost comparison with aggregated configs not an apples-to-apples comparison.
Note: Disaggregated inference configurations (e.g., MoRI SGLang, Dynamo TRTLLM) calculate input throughput per decode chip or per prefill chip, rather than per total chip count. This makes direct input throughput comparison with aggregated configs not an apples-to-apples comparison.
Note: Disaggregated inference configurations (e.g., MoRI SGLang, Dynamo TRTLLM) calculate output throughput per decode chip or per prefill chip, rather than per total chip count. This makes direct output throughput comparison with aggregated configs not an apples-to-apples comparison.
Note: Disaggregated inference configurations (e.g., MoRI SGLang, Dynamo TRTLLM) calculate power per decode chip or per prefill chip, rather than per total chip count. This makes direct power comparison with aggregated configs not an apples-to-apples comparison.
Note: Disaggregated inference configurations (e.g., MoRI SGLang, Dynamo TRTLLM) calculate Joules per decode chip or per prefill chip, rather than per total chip count. This makes direct Joules per token comparison with aggregated configs not an apples-to-apples comparison.
Matching measurements exist, but their chip series are hidden.
Shift+Scroll to zoom • Drag to pan • Double-click to reset • Click a point to pin tooltip
Qwen3.5 397B 8K / 1K Total Tokens per $1 TCO vs. Interactivity