MirrorBench - Level 1 (Autonomous Mirror Reasoning): leaderboard

MirrorBench (2026) adapts the mirror self-recognition test to multimodal models: in a simulated scene the model moves a hand to touch a mark on its own body that it can only see in a mirror. Level 1, where the mirror is announced, with no reasoning template. The score averages task success, step-wise improvement and final and peak completion ratios (at most 1; the final completion ratio is negative when the hand ends farther away than it started); higher is better.

Metric: Average score (mean of task success rate, step-wise improvement ratio, final completion ratio and peak completion ratio, each at most 1; the final completion ratio is negative when the hand ends farther from the mark than it started) at Level 1, where the mirror is announced, with no reasoning template, over both body settings, on MirrorBench's Isaac Sim mirror self-recognition scenes (7 human and robot bodies, 6 hands, 6 marks; the agent moves its hand with six directional actions to touch a mark on its own body seen in a mirror; step budget the shortest path plus 10), temperature 0; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 18 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1GPT-4.10.55#505
2Claude Sonnet 4.50.546#261
3Gemini 2.5 Pro0.465#276
4GPT-4o0.456#751
5GLM-4.5V0.364#754
6Gemini 2.5 Flash0.328#463
7Qwen 2.5 VL 32B0.261#941
8Qwen-VL-Max (MirrorBench checkpoint unspecified)0.226
9InternVL3.5-4B0.224#1930
10Qwen 2.5 VL 72B0.22#781
11LLaVA-Next-LLaMA3-8B0.051#5654
12GPT-4o Mini-0.224#1667
13Qwen 2.5 VL 7B-0.234#1862
14InternVL3.5-2B-0.296#3284
15Qwen 2.5 VL 3B-0.334#2639

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=mirrorbench-level-1-autonomous-mirror-reasoning · How It Works · Data refreshed daily, snapshot 2026-11-21.