Head-to-head AI inference benchmark comparison of GB200 NVL72 (NVIDIA Blackwell) and GB300 NVL72 (NVIDIA Blackwell) on Kimi K3 2.8T. 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 →
At 63 tok/s/user interactivity on Kimi K3 2.8T, GB200 NVL72 delivers 3674 tok/s/chip at $0.14 per million tokens; GB300 NVL72 delivers 13215 tok/s/chip at $0.05. GB300 NVL72 is 190% cheaper per token; GB300 NVL72 delivers 260% more tok/s/chip at this point.
GB200 NVL72 posts 2154 tok/s/chip for $0.24 per million tokens at 108 tok/s/user on Kimi K3 2.8T; GB300 NVL72 posts 9163 tok/s/chip for $0.07. GB300 NVL72 is 243% cheaper per token; GB300 NVL72 delivers 325% more tok/s/chip.
Throughput at 153 tok/s/user on Kimi K3 2.8T: GB200 NVL72 hits 981 tok/s/chip, GB300 NVL72 hits 6306. Per-million costs land at $0.53 and $0.10 respectively. GB300 NVL72 is 417% cheaper per token; GB300 NVL72 delivers 543% more tok/s/chip. (Numbers reflect the default agentic-traces · fp4 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | GB200 NVL72:3673.9GB300 NVL72:13215.1 | GB200 NVL72:2153.6GB300 NVL72:9163.1 | GB200 NVL72:981.4GB300 NVL72:6305.9 |
| Cost ($/M tok) | GB200 NVL72:$0.141GB300 NVL72:$0.049 | GB200 NVL72:$0.240GB300 NVL72:$0.070 | GB200 NVL72:$0.526GB300 NVL72:$0.102 |
| tok/s/MW | GB200 NVL72:1964657GB300 NVL72:6233542 | GB200 NVL72:1151643GB300 NVL72:4322231 | GB200 NVL72:524808GB300 NVL72:2974460 |
| Concurrency | GB200 NVL72:~16GB300 NVL72:~51 | GB200 NVL72:~15GB300 NVL72:~48 | GB200 NVL72:~2GB300 NVL72:~32 |
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.
