EgoBench - Static Mode - Micro Accuracy: leaderboard

Metric: MicroAcc (%; share of ground-truth tool calls issued, pooled over tasks, whole request in one user turn). Source: arxiv.org. Saturation forecast: Around June 2027. 8 models tracked.

Top models

#ModelScore
1Gemini 3.1 Pro (Preview)44.47
2Seed 2.0 Pro (Non-reasoning)40.7
3Qwen 3.5 397B A17B (Non-reasoning)36.3
4Kimi K2.5 (Non-reasoning)30.85
5GLM-5V Turbo24.24
6MiMo-V2-Omni18.09
7Qwen 3.6 Plus (Non-reasoning)18.06
8Qwen 3 VL 235B A22B Instruct16

Interactive version: theaggregate.ai/benchmark?slug=egobench-static-mode-micro-accuracy · How It Works · Data refreshed daily, snapshot 2026-09-25.