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 26–180 tok/s/user interactivity band, at 64 tok/s/user on Kimi K2.5/K2.6/K2.7-Code 1T: GB200 NVL72 runs 8980 tok/s/chip at $0.06/M tokens, GB300 NVL72 runs 9986 at $0.06/M. GB200 NVL72 is 12% cheaper per token; GB300 NVL72 delivers 11% more tok/s/chip.
Setting 103 tok/s/user as the target on Kimi K2.5/K2.6/K2.7-Code 1T, GB200 NVL72 produces 1545 tok/s/chip ($0.33 per million tokens) and GB300 NVL72 produces 1685 ($0.38). GB200 NVL72 is 14% cheaper per token; GB300 NVL72 delivers 9% more tok/s/chip.
At 141 tok/s/user interactivity on Kimi K2.5/K2.6/K2.7-Code 1T, GB200 NVL72 delivers 733 tok/s/chip at $0.70 per million tokens; GB300 NVL72 delivers 771 tok/s/chip at $0.83. GB200 NVL72 is 18% cheaper per token; GB300 NVL72 delivers 5% more tok/s/chip at this point. (Numbers reflect the default 8k/1k · fp4 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | GB200 NVL72:8979.7GB300 NVL72:9986.1 | GB200 NVL72:1545.4GB300 NVL72:1684.6 | GB200 NVL72:732.9GB300 NVL72:771.5 |
| Cost ($/M tok) | GB200 NVL72:$0.058GB300 NVL72:$0.064 | GB200 NVL72:$0.334GB300 NVL72:$0.381 | GB200 NVL72:$0.705GB300 NVL72:$0.832 |
| tok/s/MW | GB200 NVL72:4801979GB300 NVL72:4710439 | GB200 NVL72:826425GB300 NVL72:794621 | GB200 NVL72:391932GB300 NVL72:363913 |
| Concurrency | GB200 NVL72:~1203GB300 NVL72:~1146 | GB200 NVL72:~167GB300 NVL72:~68 | GB200 NVL72:~27GB300 NVL72:~27 |
Agentic inference metrics from the AgentX scenario and fixed-sequence inference metrics across models, hardware configurations, and serving parameters.
TCO $/chip/hr:
Source: SemiAnalysis Market July 2026 Pricing Surveys & AI Cloud TCO Model
Note: Disaggregated inference configurations (e.g., MoRI SGLang, Dynamo TRTLLM) calculate cost per million tokens per decode chip or per prefill chip, rather than per total chip count. This makes direct token cost comparison with aggregated configs not an apples-to-apples comparison.
Note: Disaggregated inference configurations (e.g., MoRI SGLang, Dynamo TRTLLM) calculate input throughput per decode chip or per prefill chip, rather than per total chip count. This makes direct input throughput comparison with aggregated configs not an apples-to-apples comparison.
Note: Disaggregated inference configurations (e.g., MoRI SGLang, Dynamo TRTLLM) calculate output throughput per decode chip or per prefill chip, rather than per total chip count. This makes direct output throughput comparison with aggregated configs not an apples-to-apples comparison.
Note: Disaggregated inference configurations (e.g., MoRI SGLang, Dynamo TRTLLM) calculate power per decode chip or per prefill chip, rather than per total chip count. This makes direct power comparison with aggregated configs not an apples-to-apples comparison.
Note: Disaggregated inference configurations (e.g., MoRI SGLang, Dynamo TRTLLM) calculate Joules per decode chip or per prefill chip, rather than per total chip count. This makes direct Joules per token comparison with aggregated configs not an apples-to-apples comparison.
Matching measurements exist, but their chip series are hidden.
Shift+Scroll to zoom • Drag to pan • Double-click to reset • Click a point to pin tooltip
Kimi K2.5/2.6/2.7-Code 1T 8K / 1K Total Tokens per $1 TCO vs. Interactivity