PerceptRL Single-turn - RealWorldCounting: 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: Not forecast. 16 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Doubao Seed 2.0 Mini | 37.9 | #160 |
| 2 | Qwen 3.5 Plus | 35.4 | #131 |
| 3 | Gemini 3 Flash | 34.6 | #103 |
| 4 | Qwen 3.6 35B A3B | 24.3 | #233 |
| 5 | Qwen 3.5 Flash | 22 | #235 |
| 6 | Gemini 2.5 Pro | 19.2 | #156 |
| 7 | Grok 4.5 | 19 | #53 |
| 8 | Claude Opus 4.8 | 17.8 | #39 |
| 9 | Grok 4.3 | 14.9 | #196 |
| 10 | Claude Opus 4.6 | 13.8 | #65 |
| 11 | Kimi K2.5 | 13.4 | #148 |
| 12 | Claude Opus 4.7 | 13.4 | #51 |
| 13 | Kimi K2.6 | 12.9 | #111 |
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-realworldcounting · How It Works · Data refreshed daily, snapshot 2026-10-29.