Kimi K2.5/K2.6 1T — B200 vs H200
Head-to-head AI inference benchmark comparison of B200 (NVIDIA Blackwell) and H200 (NVIDIA Hopper) 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.
Near the low end of the 37–106 tok/s/user interactivity band, at 54 tok/s/user on Kimi K2.5/K2.6 1T: B200 runs 375 tok/s/GPU at $1.44/M tokens, H200 runs 295 at $1.31/M. H200 is 10% cheaper per token; B200 delivers 27% more tok/s/GPU.
Setting 71 tok/s/user as the target on Kimi K2.5/K2.6 1T, B200 produces 254 tok/s/GPU ($2.13 per million tokens) and H200 produces 215 ($1.80). H200 is 18% cheaper per token; B200 delivers 18% more tok/s/GPU.
At 89 tok/s/user interactivity on Kimi K2.5/K2.6 1T, B200 delivers 181 tok/s/GPU at $2.98 per million tokens; H200 delivers 154 tok/s/GPU at $2.56. H200 is 17% cheaper per token; B200 delivers 18% more tok/s/GPU at this point. (Numbers reflect the default 1k/1k · int4 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | B200:375.4H200:295.4 | B200:253.7H200:214.6 | B200:181.5H200:154.3 |
| Cost ($/M tok) | B200:$1.444H200:$1.313 | B200:$2.126H200:$1.804 | B200:$2.985H200:$2.560 |
| tok/s/MW | B200:172987H200:170736 | B200:116921H200:124042 | B200:83638H200:89206 |
| Concurrency | B200:~29H200:~22 | B200:~14H200:~12 | B200:~8H200:~7 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.