B200: FP4 vs INT4 Precision Comparison

FP4 posts 8817 tok/s/chip for $0.05 per million tokens at 49 tok/s/user on Kimi K2.5/K2.6/K2.7-Code 1T (B200); INT4 posts 1459 tok/s/chip for $0.33. FP4 is 504% cheaper per token; FP4 delivers 504% more tok/s/chip. Quantization-level accuracy differences are tracked on the evaluation tab.

Throughput at 68 tok/s/user on Kimi K2.5/K2.6/K2.7-Code 1T (B200): FP4 hits 3630 tok/s/chip, INT4 hits 1016. Per-million costs land at $0.13 and $0.47 respectively. FP4 is 257% cheaper per token; FP4 delivers 257% more tok/s/chip. The cost-throughput tradeoff from lower precision is only part of the picture — see the evaluation page for accuracy data.

Toward the upper edge of the 30–105 tok/s/user interactivity band, at 87 tok/s/user on Kimi K2.5/K2.6/K2.7-Code 1T (B200): FP4 runs 1968 tok/s/chip at $0.24/M tokens, INT4 runs 655 at $0.73/M. FP4 is 201% cheaper per token; FP4 delivers 201% more tok/s/chip. Precision changes affect both inference speed and model quality — consult the evaluation tab for accuracy benchmarks. (Numbers reflect the default 8k/1k selection for this URL — table and chart below update if you change sequence or model in the controls. Each side uses the best available serving configuration for that precision, which may include speculative decoding such as MTP where recipes exist — the same convention as the other comparison pages.)