Kimi K2.5/K2.6/K2.7-Code 1T · B300 · Precision comparison

B300: FP4 vs INT4 precision comparison

How FP4 and INT4 precision affect Kimi K2.5/K2.6/K2.7-Code 1T inference on B300 (NVIDIA Blackwell). Throughput, latency, and cost across LLM workloads. Use the chart controls below to switch sequences and metrics — same interactions as the main inference chart.

Throughput at 44 tok/s/user on Kimi K2.5/K2.6/K2.7-Code 1T (B300): FP4 hits 3,878 tok/s/chip, INT4 hits 2,025. Costs per million tokens land at $0.16 and $0.31 respectively. FP4 delivers 91% more tok/s/chip on the same chip, which makes it 48% cheaper per token. The cost-throughput tradeoff from lower precision is only part of the picture — see the evaluation page for accuracy data.

Around the middle of the 23–108 tok/s/user interactivity band, at 65 tok/s/user on Kimi K2.5/K2.6/K2.7-Code 1T (B300): FP4 runs 2,818 tok/s/chip at $0.22 per million tokens, INT4 runs 1,409 at $0.45. FP4 delivers 100% more tok/s/chip on the same chip, which makes it 50% cheaper per token. Precision changes affect both inference speed and model quality — consult the evaluation tab for accuracy benchmarks.

At 87 tok/s/user on Kimi K2.5/K2.6/K2.7-Code 1T (B300), FP4 delivers 2,079 tok/s/chip at $0.30 per million tokens; INT4 delivers 882 tok/s/chip at $0.71. FP4 delivers 136% more tok/s/chip on the same chip, which makes it 58% cheaper per token. Lower-precision quantization trades model accuracy for throughput — check the evaluation page for quality impact. (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.)

Interpolated from real benchmark data. Edit target interactivity values below to compare at different operating points.

Metric
Interactivity (tok/s/user)
Interactivity (tok/s/user)
Interactivity (tok/s/user)
Throughput (tok/s/chip)
FP43,878INT42,025
FP42,818INT41,409
FP42,079INT4882
Cost ($/M tok)
FP4$0.16INT4$0.31
FP4$0.22INT4$0.45
FP4$0.30INT4$0.71
tok/s/MW
FP42,040,825INT41,065,851
FP41,483,089INT4741,740
FP41,094,062INT4464,463
Concurrency
FP4~40INT4~20
FP4~20INT4~10
FP4~11INT4~4
Kimi K2.5/K2.6/K2.7-Code 1T: FP4 versus INT4 throughput and cost at matched interactivity levels on B300
FP4 versus INT4 throughput and cost per million tokens on B300 for this comparison's canonical default workload
Open full-size image ↗