B300: FP4 vs FP8 Precision Comparison

FP4 posts 1775 tok/s/GPU for $0.37 per million tokens at 51 tok/s/user on GLM 5/5.1 (B300); FP8 posts 671 tok/s/GPU for $0.96. FP4 is 162% cheaper per token; FP4 delivers 164% more tok/s/GPU. Quantization-level accuracy differences are tracked on the evaluation tab.

Throughput at 80 tok/s/user on GLM 5/5.1 (B300): FP4 hits 1034 tok/s/GPU, FP8 hits 424. Per-million costs land at $0.63 and $1.52 respectively. FP4 is 141% cheaper per token; FP4 delivers 144% 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 24–136 tok/s/user interactivity band, at 108 tok/s/user on GLM 5/5.1 (B300): FP4 runs 563 tok/s/GPU at $1.16/M tokens, FP8 runs 235 at $2.79/M. FP4 is 139% cheaper per token; FP4 delivers 139% 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.)