Kimi K2.5/K2.6/K2.7-Code 1T — GB200 NVL72 vs GB300 NVL72
Head-to-head AI inference benchmark comparison of GB200 NVL72 (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.
Near the low end of the 18–153 tok/s/user interactivity band, at 52 tok/s/user on Kimi K2.5/K2.6/K2.7-Code 1T: GB200 NVL72 runs 8760 tok/s/GPU at $0.07/M tokens, GB300 NVL72 runs 4381 at $0.16/M. GB200 NVL72 is 133% cheaper per token; GB200 NVL72 delivers 100% more tok/s/GPU.
Setting 86 tok/s/user as the target on Kimi K2.5/K2.6/K2.7-Code 1T, GB200 NVL72 produces 1614 tok/s/GPU ($0.38 per million tokens) and GB300 NVL72 produces 461 ($1.56). GB200 NVL72 is 307% cheaper per token; GB200 NVL72 delivers 250% more tok/s/GPU.
At 120 tok/s/user interactivity on Kimi K2.5/K2.6/K2.7-Code 1T, GB200 NVL72 delivers 316 tok/s/GPU at $1.92 per million tokens; GB300 NVL72 delivers 228 tok/s/GPU at $3.22. GB200 NVL72 is 68% cheaper per token; GB200 NVL72 delivers 38% more tok/s/GPU at this point. (Numbers reflect the default 1k/1k · fp4 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | GB200 NVL72:8760.0GB300 NVL72:4381.5 | GB200 NVL72:1614.0GB300 NVL72:460.9 | GB200 NVL72:315.6GB300 NVL72:228.5 |
| Cost ($/M tok) | GB200 NVL72:$0.070GB300 NVL72:$0.163 | GB200 NVL72:$0.382GB300 NVL72:$1.557 | GB200 NVL72:$1.921GB300 NVL72:$3.219 |
| tok/s/MW | GB200 NVL72:4171441GB300 NVL72:2086405 | GB200 NVL72:768567GB300 NVL72:219485 | GB200 NVL72:150275GB300 NVL72:108793 |
| Concurrency | GB200 NVL72:~4255GB300 NVL72:~1198 | GB200 NVL72:~462GB300 NVL72:~92 | GB200 NVL72:~47GB300 NVL72:~33 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.