Kimi K3 2.8T — B200 vs GB300 NVL72
Head-to-head AI inference benchmark comparison of B200 (NVIDIA Blackwell) and GB300 NVL72 (NVIDIA Blackwell) on Kimi K3 2.8T. 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.
AgentX replays real coding-agent sessions rather than fixed-length prompts, so context grows turn over turn and most of each request is served from cache instead of being recomputed. That turns the comparison into a systems question: KV transfer between nodes, prefix-aware routing, and cache capacity all move the curve alongside raw chip throughput. Fixed-sequence workloads stay the clean baseline for kernel and silicon performance, so the two scenarios answer different questions about the same hardware. Learn more about AgentX →
Near the low end of the 7–207 tok/s/user interactivity band, at 57 tok/s/user on Kimi K3 2.8T: B200 runs 4011 tok/s/chip at $0.12/M tokens, GB300 NVL72 runs 7833 at $0.08/M. GB300 NVL72 is 46% cheaper per token; GB300 NVL72 delivers 95% more tok/s/chip.
Setting 107 tok/s/user as the target on Kimi K3 2.8T, B200 produces 1996 tok/s/chip ($0.24 per million tokens) and GB300 NVL72 produces 5044 ($0.13). GB300 NVL72 is 89% cheaper per token; GB300 NVL72 delivers 153% more tok/s/chip.
At 157 tok/s/user interactivity on Kimi K3 2.8T, B200 delivers 1288 tok/s/chip at $0.37 per million tokens; GB300 NVL72 delivers 3305 tok/s/chip at $0.19. GB300 NVL72 is 92% cheaper per token; GB300 NVL72 delivers 157% more tok/s/chip at this point. (Numbers reflect the default agentic-traces · fp4 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | B200:4011.3GB300 NVL72:7833.0 | B200:1996.4GB300 NVL72:5044.3 | B200:1288.2GB300 NVL72:3305.1 |
| Cost ($/M tok) | B200:$0.120GB300 NVL72:$0.082 | B200:$0.241GB300 NVL72:$0.127 | B200:$0.373GB300 NVL72:$0.194 |
| tok/s/MW | B200:2345806GB300 NVL72:3694809 | B200:1167470GB300 NVL72:2379408 | B200:753307GB300 NVL72:1559006 |
| Concurrency | B200:~14GB300 NVL72:~63 | B200:~6GB300 NVL72:~34 | B200:~3GB300 NVL72:~7 |