Kimi K2.5/K2.6 1T — B200 vs GB200 NVL72
Head-to-head AI inference benchmark comparison of B200 (NVIDIA Blackwell) and GB200 NVL72 (NVIDIA Blackwell) on Kimi K2.5/K2.6 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 61 tok/s/user on Kimi K2.5/K2.6 1T: B200 hits 1020 tok/s/GPU, GB200 NVL72 hits 1797. Per-million costs land at $0.53 and $0.34 respectively. GB200 NVL72 is 56% cheaper per token; GB200 NVL72 delivers 76% more tok/s/GPU.
B200 / GB200 NVL72 on Kimi K2.5/K2.6 1T at 88 tok/s/user: 522 / 781 tok/s/GPU, $1.00 / $0.79 per million tokens. GB200 NVL72 is 27% cheaper per token; GB200 NVL72 delivers 50% more tok/s/GPU.
Toward the upper edge of the 35–140 tok/s/user interactivity band, at 114 tok/s/user on Kimi K2.5/K2.6 1T: B200 runs 344 tok/s/GPU at $1.56/M tokens, GB200 NVL72 runs 270 at $2.28/M. B200 is 46% cheaper per token; B200 delivers 28% more tok/s/GPU. (Numbers reflect the default 1k/1k · fp4 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | B200:1020.3GB200 NVL72:1796.7 | B200:522.3GB200 NVL72:781.2 | B200:343.9GB200 NVL72:269.7 |
| Cost ($/M tok) | B200:$0.529GB200 NVL72:$0.339 | B200:$1.003GB200 NVL72:$0.788 | B200:$1.558GB200 NVL72:$2.276 |
| tok/s/MW | B200:470171GB200 NVL72:855574 | B200:240687GB200 NVL72:372007 | B200:158488GB200 NVL72:128422 |
| Concurrency | B200:~35GB200 NVL72:~667 | B200:~12GB200 NVL72:~173 | B200:~6GB200 NVL72:~30 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.