B200: FP4 vs FP8 Precision Comparison

At 46 tok/s/user on GLM 5/5.1 (B200), FP4 delivers 2935 tok/s/chip at $0.16 per million tokens; FP8 delivers 1381 tok/s/chip at $0.35. FP4 is 113% cheaper per token; FP4 delivers 113% more tok/s/chip. Lower-precision quantization trades model accuracy for throughput — check the evaluation page for quality impact.

FP4 posts 1980 tok/s/chip for $0.24 per million tokens at 78 tok/s/user on GLM 5/5.1 (B200); FP8 posts 906 tok/s/chip for $0.53. FP4 is 119% cheaper per token; FP4 delivers 119% more tok/s/chip. Quantization-level accuracy differences are tracked on the evaluation tab.

Throughput at 110 tok/s/user on GLM 5/5.1 (B200): FP4 hits 1417 tok/s/chip, FP8 hits 569. Per-million costs land at $0.34 and $0.84 respectively. FP4 is 149% cheaper per token; FP4 delivers 149% more tok/s/chip. The cost-throughput tradeoff from lower precision is only part of the picture — see the evaluation page for accuracy data. (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.)