Kimi K3 2.8T — B200 vs B300
Head-to-head AI inference benchmark comparison of B200 (NVIDIA Blackwell) and B300 (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 →
Throughput at 57 tok/s/user on Kimi K3 2.8T: B200 hits 4011 tok/s/chip, B300 hits 5615. Per-million costs land at $0.12 and $0.11 respectively. B300 is 7% cheaper per token; B300 delivers 40% more tok/s/chip.
B200 / B300 on Kimi K3 2.8T at 105 tok/s/user: 2036 / 3911 tok/s/chip, $0.24 / $0.16 per million tokens. B300 is 47% cheaper per token; B300 delivers 92% more tok/s/chip.
Toward the upper edge of the 8–202 tok/s/user interactivity band, at 154 tok/s/user on Kimi K3 2.8T: B200 runs 1320 tok/s/chip at $0.36/M tokens, B300 runs 2069 at $0.30/M. B300 is 20% cheaper per token; B300 delivers 57% more tok/s/chip. (Numbers reflect the default agentic-traces · fp4 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | B200:4011.3B300:5615.2 | B200:2036.2B300:3911.3 | B200:1319.9B300:2068.8 |
| Cost ($/M tok) | B200:$0.120B300:$0.112 | B200:$0.236B300:$0.161 | B200:$0.364B300:$0.303 |
| tok/s/MW | B200:2345806B300:2955360 | B200:1190752B300:2058569 | B200:771898B300:1088859 |
| Concurrency | B200:~14B300:~9 | B200:~6B300:~5 | B200:~3B300:~2 |