MirrorBench - Level 2 (Implicit Mirror Discovery): 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 2, where the mirror is not mentioned. The score averages task success, step-wise improvement and final and peak completion ratios (at most 1; 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 2, where the mirror is not mentioned, 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
1Gemini 2.5 Pro0.256#276
2Gemini 2.5 Flash0.255#463
3Qwen-VL-Max (MirrorBench checkpoint unspecified)0.175
4GPT-4o0.129#751
5Qwen 2.5 VL 32B0.112#941
6LLaVA-Next-LLaMA3-8B0.111#5654
7Qwen 2.5 VL 72B0.058#781
8InternVL3.5-4B0.019#1930
9GPT-4o Mini-0.031#1667
10LLaVA-1.6-7B-0.077#4955
11GPT-4.1-0.088#505
12Qwen 2.5 VL 7B-0.126#1862
13Claude Sonnet 4.5-0.175#261
14GLM-4.5V-0.218#754
15InternVL3.5-2B-0.239#3284

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-2-implicit-mirror-discovery · How It Works · Data refreshed daily, snapshot 2026-11-21.