PerceptRL Single-turn - TreeBench: leaderboard
PerceptRL tests visual understanding and spatial reasoning, with separate results for single-turn answers and agents using tools. Accuracy (%).
Metric: Accuracy (%). Source: arxiv.org. Saturation forecast: Around 2029. 16 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3.5 Plus | 62.2 | #131 |
| 2 | Grok 4.5 | 59.5 | #53 |
| 3 | Qwen 3.5 Flash | 58.8 | #235 |
| 4 | Gemini 3 Flash | 58.3 | #103 |
| 5 | Claude Opus 4.8 | 58.3 | #39 |
| 6 | Qwen 3.6 35B A3B | 57.8 | #233 |
| 7 | Claude Opus 4.7 | 57.3 | #51 |
| 8 | Doubao Seed 2.0 Mini | 56.3 | #160 |
| 9 | Kimi K2.6 | 55.6 | #111 |
| 10 | Gemini 2.5 Pro | 54.8 | #156 |
| 11 | Claude Opus 4.6 | 53.3 | #65 |
| 12 | Grok 4.3 | 53.1 | #196 |
| 13 | Kimi K2.5 | 52.6 | #148 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=perceptrl-single-turn-treebench · How It Works · Data refreshed daily, snapshot 2026-10-29.