Kimi K2.5/K2.6 1T — B300 vs GB200 NVL72
Head-to-head AI inference benchmark comparison of B300 (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.
B300 / GB200 NVL72 on Kimi K2.5/K2.6 1T at 69 tok/s/user: 932 / 1370 tok/s/GPU, $0.70 / $0.45 per million tokens. GB200 NVL72 is 56% cheaper per token; GB200 NVL72 delivers 47% more tok/s/GPU.
Around the middle of the 35–172 tok/s/user interactivity band, at 104 tok/s/user on Kimi K2.5/K2.6 1T: B300 runs 899 tok/s/GPU at $0.72/M tokens, GB200 NVL72 runs 570 at $1.22/M. B300 is 68% cheaper per token; B300 delivers 58% more tok/s/GPU.
Setting 138 tok/s/user as the target on Kimi K2.5/K2.6 1T, B300 produces 493 tok/s/GPU ($1.32 per million tokens) and GB200 NVL72 produces 151 ($4.11). B300 is 211% cheaper per token; B300 delivers 227% more tok/s/GPU. (Numbers reflect the default 1k/1k · fp4 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | B300:931.5GB200 NVL72:1370.0 | B300:899.5GB200 NVL72:570.0 | B300:493.4GB200 NVL72:150.7 |
| Cost ($/M tok) | B300:$0.698GB200 NVL72:$0.448 | B300:$0.723GB200 NVL72:$1.218 | B300:$1.323GB200 NVL72:$4.114 |
| tok/s/MW | B300:429274GB200 NVL72:652400 | B300:414503GB200 NVL72:271415 | B300:227387GB200 NVL72:71768 |
| Concurrency | B300:~28GB200 NVL72:~430 | B300:~19GB200 NVL72:~112 | B300:~7GB200 NVL72:~14 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.