Kimi K2.5/K2.6/K2.7-Code 1T — B200 vs GB300 NVL72
Head-to-head AI inference benchmark comparison of B200 (NVIDIA Blackwell) and GB300 NVL72 (NVIDIA Blackwell) on Kimi K2.5/K2.6/K2.7-Code 1T. 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 63 tok/s/user on Kimi K2.5/K2.6/K2.7-Code 1T: B200 hits 3774 tok/s/chip, GB300 NVL72 hits 10203. Per-million costs land at $0.13 and $0.06 respectively. GB300 NVL72 is 102% cheaper per token; GB300 NVL72 delivers 170% more tok/s/chip.
B200 / GB300 NVL72 on Kimi K2.5/K2.6/K2.7-Code 1T at 103 tok/s/user: 1807 / 1685 tok/s/chip, $0.27 / $0.38 per million tokens. B200 is 41% cheaper per token; B200 delivers 7% more tok/s/chip.
Toward the upper edge of the 24–182 tok/s/user interactivity band, at 143 tok/s/user on Kimi K2.5/K2.6/K2.7-Code 1T: B200 runs 585 tok/s/chip at $0.82/M tokens, GB300 NVL72 runs 731 at $0.87/M. B200 is 6% cheaper per token; GB300 NVL72 delivers 25% more tok/s/chip. (Numbers reflect the default 8k/1k · fp4 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | B200:3773.9GB300 NVL72:10202.8 | B200:1807.1GB300 NVL72:1684.6 | B200:585.1GB300 NVL72:731.1 |
| Cost ($/M tok) | B200:$0.127GB300 NVL72:$0.063 | B200:$0.270GB300 NVL72:$0.382 | B200:$0.822GB300 NVL72:$0.869 |
| tok/s/MW | B200:2206960GB300 NVL72:4812653 | B200:1056807GB300 NVL72:794621 | B200:342142GB300 NVL72:344856 |
| Concurrency | B200:~231GB300 NVL72:~1202 | B200:~77GB300 NVL72:~68 | B200:~10GB300 NVL72:~26 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.