Kimi K2 0905 — benchmark results

September 5, 2025 update of Moonshot's open Kimi K2 MoE with improved agentic coding and a 256K context. Provider: Moonshot. Released 2025-09-05. Access: Open.

Unified ELO 1685 ± 17, rank #301 of 1776 rated models, from 201 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
AGC-Bench - future_ideas2.07Dataset z-score100
AGC-Bench - hypogen2.19Dataset z-score100
AGC-Bench - pron_vs_prompt1.41Dataset z-score100
AGC-Bench - thenextchapter2.77Dataset z-score100
Konkur 1404 - Art57.47Accuracy (%, text-only)100
AGC-Bench - conceptual_design2.02Dataset z-score98.8
AGC-Bench - cpers1.4Dataset z-score98.8
AGC-Bench - cue_word_story1.16Dataset z-score98.8
AGC-Bench - tinystories1.67Dataset z-score98.8
AGC-Bench - rpgbench1.54Dataset z-score97.5
AGC-Bench - tinyfabulist1.45Dataset z-score95.1
Konkur 1404 - Mathematics52.17Accuracy (%, text-only)94.7

Interactive version: theaggregate.ai/model?slug=kimi-k2-0905 · How the rankings work · Data refreshed daily, snapshot 2026-07-22.