Kimi K2.5/K2.6/K2.7-Code 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/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.
B200 / H200 on Kimi K2.5/K2.6/K2.7-Code 1T at 54 tok/s/user: 353 / 294 tok/s/chip, $1.36 / $1.13 per million tokens. H200 is 20% cheaper per token; B200 delivers 20% more tok/s/chip.
Around the middle of the 37–105 tok/s/user interactivity band, at 71 tok/s/user on Kimi K2.5/K2.6/K2.7-Code 1T: B200 runs 236 tok/s/chip at $2.02/M tokens, H200 runs 221 at $1.53/M. H200 is 32% cheaper per token; B200 delivers 7% more tok/s/chip.
Setting 88 tok/s/user as the target on Kimi K2.5/K2.6/K2.7-Code 1T, B200 produces 168 tok/s/chip ($2.86 per million tokens) and H200 produces 158 ($2.16). H200 is 32% cheaper per token; B200 delivers 6% more tok/s/chip. (Numbers reflect the default 1k/1k · int4 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | B200:353.2H200:294.0 | B200:236.4H200:220.6 | B200:168.3H200:158.1 |
| Cost ($/M tok) | B200:$1.358H200:$1.130 | B200:$2.018H200:$1.526 | B200:$2.856H200:$2.157 |
| tok/s/MW | B200:206546H200:214590 | B200:138251H200:161039 | B200:98439H200:115388 |
| Concurrency | B200:~26H200:~22 | B200:~13H200:~12 | B200:~8H200:~7 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.