B200: FP4 vs FP8 Precision Comparison

Near the low end of the 15–383 tok/s/user interactivity band, at 106 tok/s/user on MiniMax M3 428B (B200): FP4 runs 2079 tok/s/chip at $0.23/M tokens, FP8 runs 1440 at $0.33/M. FP4 is 44% cheaper per token; FP4 delivers 44% more tok/s/chip. Precision changes affect both inference speed and model quality — consult the evaluation tab for accuracy benchmarks.

At 199 tok/s/user on MiniMax M3 428B (B200), FP4 delivers 861 tok/s/chip at $0.56 per million tokens; FP8 delivers 554 tok/s/chip at $0.87. FP4 is 55% cheaper per token; FP4 delivers 55% more tok/s/chip. Lower-precision quantization trades model accuracy for throughput — check the evaluation page for quality impact.

FP4 posts 291 tok/s/chip for $1.65 per million tokens at 291 tok/s/user on MiniMax M3 428B (B200); FP8 posts 195 tok/s/chip for $2.47. FP4 is 49% cheaper per token; FP4 delivers 49% more tok/s/chip. Quantization-level accuracy differences are tracked on the evaluation tab. (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.)