MirrorBench - Human Body: 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. Score with anthropomorphic bodies and hands, over four prompt levels. 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) with the anthropomorphic bodies and hands, over the four prompt levels, 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
1Gemini 2.5 Pro0.47#280
2Gemini 2.5 Flash0.409#466
3GPT-4.10.356#506
4GPT-4o0.346#754
5Qwen-VL-Max (MirrorBench checkpoint unspecified)0.336
6Claude Sonnet 4.50.308#267
7Qwen 2.5 VL 72B0.224#782
8GLM-4.5V0.193#756
9Qwen 2.5 VL 32B0.142#944
10InternVL3.5-4B0.051#1946
11GPT-4o Mini-0.062#1680
12LLaVA-Next-LLaMA3-8B-0.18#5702
13LLaVA-1.6-7B-0.241#4990
14Qwen 2.5 VL 7B-0.247#1878
15InternVL3.5-1B-0.253#6140

No result here: #2 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=mirrorbench-human-body · How It Works · Data refreshed daily, snapshot 2026-11-22.