AgentRoboCritic - Overall Accuracy: leaderboard
AgentRoboCritic (arXiv:2611.26613v1, Table 1): zero-shot multimodal action-level critic evaluation on 400 human-annotated direct robot-control actions from 12 manipulation benchmarks. Overall accuracy (%) requiring both action outcome and failure mode correct (off-the-shelf models only); higher is better.
Metric: Overall Accuracy (%). Source: arxiv.org. Saturation forecast: Around May 2027. 10 models tracked. Updated 21 Nov 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-6 Astra | 71.25 | #10 |
| 2 | Claude Fable 5.2 | 71 | #1 |
| 3 | GPT-6.1 Sol | 69.75 | #15 |
| 4 | GPT-6.1 Terra | 66.75 | #25 |
| 5 | Claude Opus 5.5 | 66 | #9 |
| 6 | GLM-5.3 Flash | 58.25 | #133 |
| 7 | Kimi K3.1 Pro | 52.5 | #36 |
| 8 | DeepSeek V4.1 Flash (1017) | 52 | #47 |
| 9 | Claude Sonnet 5.5 | 50.75 | #24 |
| 10 | Muse Glimmer 30B | 34.75 | #398 |
No result here: #3 Gemini 4 Neon, #4 GPT-6.2 Astra, #7 Gemini 4 Carbon.
Interactive version: theaggregate.ai/benchmark?slug=agentrobocritic-overall-accuracy · How It Works · Data refreshed daily, snapshot 2026-11-21.