KAT-Coder-Pro V2.5: benchmark results

Provider: Kuaishou. Released 2026-07-10. Access: API.

Unified ELO 1840 ± 23, rank #100 of 2656 rated models, from 11 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
BenchmarkList ECI134.57Capability Index (ECI)84.2
AI Chess Leaderboard (Reasoning)718Elo55.1
ObviousBench90.97Answer pass³ (%)50.4
Gert Labs Rankings37.67GScore (%)44.4
Tinybird AI SQL Benchmark - First-Attempt Success Rate92Questions answered with a valid query on the first attempt (40.5
Tinybird AI SQL Benchmark - Success Rate98Questions answered with a valid query within 3 retries (%)35.5
KiloBench - Terminal Bench 2.050.34Task completion in Kilo's agent harness, mean of 5 attempts 33.3
Tinybird AI SQL Benchmark - Exactness43.99Result exactness vs human reference queries (0-100)30.7
SvelteBench82.2Average pass@1 (%)30.6
LM Market Cap LMC Score40LMC Score (0-100)28
AI Chess Leaderboard (Continuation)466Elo26.3

Interactive version: theaggregate.ai/model?slug=kat-coder-pro-v2-5 · How It Works · Data refreshed daily, snapshot 2026-09-19.