B300: FP4 vs BF16 Precision Comparison

At 102 tok/s/user on Qwen 3.5 397B-A17B (B300), FP4 delivers 4220 tok/s/GPU at $0.15 per million tokens; BF16 delivers 1366 tok/s/GPU at $0.47. FP4 is 211% cheaper per token; FP4 delivers 209% more tok/s/GPU. Lower-precision quantization trades model accuracy for throughput — check the evaluation page for quality impact.

FP4 posts 2057 tok/s/GPU for $0.32 per million tokens at 160 tok/s/user on Qwen 3.5 397B-A17B (B300); BF16 posts 712 tok/s/GPU for $0.91. FP4 is 188% cheaper per token; FP4 delivers 189% more tok/s/GPU. Quantization-level accuracy differences are tracked on the evaluation tab.

Throughput at 218 tok/s/user on Qwen 3.5 397B-A17B (B300): FP4 hits 1242 tok/s/GPU, BF16 hits 466. Per-million costs land at $0.52 and $1.37 respectively. FP4 is 161% cheaper per token; FP4 delivers 167% more tok/s/GPU. The cost-throughput tradeoff from lower precision is only part of the picture — see the evaluation page for accuracy data. (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.)