KAT-Coder-Pro V1 — benchmark results

Kwaipilot's (Kuaishou) first production agentic-coding KAT-Coder API model. Provider: Kuaishou. Released 2025-10-01. Access: API.

Unified ELO 1701 ± 35, rank #268 of 1776 rated models, from 60 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
AA Long Context Reasoning74Accuracy (%)98.3
AA-LCR74Score (self-reported)98.3
AA AIME 202594.67Accuracy (%)96.1
AA Humanity's Last Exam33.41Accuracy (%)92.1
Wolfram LLM Benchmarking Project58.9Correct Functionality (%)88.3
AA LiveCodeBench74.71Pass@1 (%)86.1
AA TAU-2 Bench88.6Accuracy (%)82.8
AA IFBench68.37Accuracy (%)81.9
AA MMLU-Pro81.27Accuracy (%)73.7
Artificial Analysis Intelligence Index28.26Intelligence Index72.5
AA Global-MMLU-Lite - Chinese88.42Accuracy (%)71
AA Omniscience - Software Engineering (SWE) - Julia24Accuracy (%)70.4

Interactive version: theaggregate.ai/model?slug=kat-coder-pro-v1 · How the rankings work · Data refreshed daily, snapshot 2026-07-22.