B300: FP4 vs FP8 Precision Comparison

FP4 posts 15523 tok/s/GPU for $0.04 per million tokens at 47 tok/s/user on MiniMax M2.5/M2.7 (B300); FP8 posts 4923 tok/s/GPU for $0.13. FP4 is 215% cheaper per token; FP4 delivers 215% more tok/s/GPU. Quantization-level accuracy differences are tracked on the evaluation tab.

Throughput at 79 tok/s/user on MiniMax M2.5/M2.7 (B300): FP4 hits 6656 tok/s/GPU, FP8 hits 1534. Per-million costs land at $0.10 and $0.43 respectively. FP4 is 334% cheaper per token; FP4 delivers 334% 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 16–142 tok/s/user interactivity band, at 111 tok/s/user on MiniMax M2.5/M2.7 (B300): FP4 runs 2536 tok/s/GPU at $0.26/M tokens, FP8 runs 823 at $0.79/M. FP4 is 209% cheaper per token; FP4 delivers 208% 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.)