FREAK - Attribute: leaderboard
Metric: Accuracy (%) on the attribute questions of FREAK's 1,799 questions (1,000 multiple-choice with cyclic option permutations and 799 free-form judged by GPT-5-mini) on photorealistic images edited to contradict commonsense; reasoning enabled where the model supports it; higher is better. Source: arxiv.org. Saturation forecast: Around 2029. 18 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Pro | 56.12 | #145 |
| 2 | O3 (High) | 49.89 | #121 (O3) |
| 3 | InternVL3-78B | 48.49 | #345 |
| 4 | GPT-4.1 | 48.24 | #240 |
| 5 | Gemini 2.5 Flash | 48.02 | #237 |
| 6 | GLM-4.5V | 47.89 | #339 |
| 7 | O4 Mini | 47.22 | #172 |
| 8 | Qwen 2.5 VL 72B Instruct | 46.58 | #364 |
| 9 | InternVL3-38B | 46.46 | #395 |
| 10 | Qwen 2.5 VL 32B Instruct | 42.63 | #443 |
| 11 | Phi-4 Multimodal Instruct | 36.6 | #896 |
| 12 | Claude Sonnet 4 (Thinking) | 33.95 | #194 (Claude Sonnet 4) |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=freak-attribute · How It Works · Data refreshed daily, snapshot 2026-10-11.