B200: FP4 vs INT4 Precision Comparison

FP4 posts 1057 tok/s/GPU for $0.51 per million tokens at 55 tok/s/user on Kimi K2.5/K2.6/K2.7-Code 1T (B200); INT4 posts 344 tok/s/GPU for $1.57. FP4 is 208% cheaper per token; FP4 delivers 207% more tok/s/GPU. Quantization-level accuracy differences are tracked on the evaluation tab.

Throughput at 73 tok/s/user on Kimi K2.5/K2.6/K2.7-Code 1T (B200): FP4 hits 713 tok/s/GPU, INT4 hits 227. Per-million costs land at $0.76 and $2.37 respectively. FP4 is 213% cheaper per token; FP4 delivers 215% more tok/s/GPU. 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 37–109 tok/s/user interactivity band, at 91 tok/s/user on Kimi K2.5/K2.6/K2.7-Code 1T (B200): FP4 runs 507 tok/s/GPU at $1.06/M tokens, INT4 runs 158 at $3.45/M. FP4 is 226% cheaper per token; FP4 delivers 221% more tok/s/GPU. Precision changes affect both inference speed and model quality — consult the evaluation tab for accuracy benchmarks. (Numbers reflect the default 1k/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.)