Head-to-head AI inference benchmark comparison of B300 (NVIDIA Blackwell) and H200 (NVIDIA Hopper) on DeepSeek V4.1 Flash 552B. Latency, throughput, and cost across LLM workloads. Use the chart controls below to switch sequences, precisions, and metrics β 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 β
Setting 111 tok/s/user as the target on DeepSeek V4.1 Flash 552B, B300 produces 189594 tok/s/chip ($0.00 per million tokens) and H200 produces 15297 ($0.02). B300 is 569% cheaper per token; B300 delivers 1139% more tok/s/chip.
At 181 tok/s/user interactivity on DeepSeek V4.1 Flash 552B, B300 delivers 107348 tok/s/chip at $0.01 per million tokens; H200 delivers 9526 tok/s/chip at $0.04. B300 is 508% cheaper per token; B300 delivers 1027% more tok/s/chip at this point.
B300 posts 56351 tok/s/chip for $0.01 per million tokens at 251 tok/s/user on DeepSeek V4.1 Flash 552B; H200 posts 5386 tok/s/chip for $0.06. B300 is 465% cheaper per token; B300 delivers 946% more tok/s/chip. (Numbers reflect the default agentic-traces Β· fp4 selection for this URL β table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | B300:189594.3H200:15297.0 | B300:107348.4H200:9526.3 | B300:56351.2H200:5386.2 |
| Cost ($/M tok) | B300:$0.003H200:$0.022 | B300:$0.006H200:$0.036 | B300:$0.011H200:$0.063 |
| tok/s/MW | B300:99786495H200:11165667 | B300:56499182H200:6953529 | B300:29658525H200:3931524 |
| Concurrency | B300:~57H200:~10 | B300:~28H200:~5 | B300:~16H200:~1 |
Agentic inference metrics from the AgentX scenario and fixed-sequence inference metrics across models, hardware configurations, and serving parameters.
No measurements to plot for this selection. Review the benchmark controls above or adjust quick filters.
