VisReason (Vision-Centric) - Fill-in-the-Blank: leaderboard
Metric: Accuracy (%) on the 441 fill-in-the-blank questions (answer extracted by regular expression) of VisReason, pooled over the 10 reasoning categories, zero-shot with chain-of-thought prompt templates, one per answer format; rows not shaded gray in the paper use explicit reasoning (GPT-5 and Gemini 3 Pro at low reasoning effort); higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 20 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3 Pro (Preview) (Low) | 47.4 |
| 2 | GPT-5.2 (Thinking) | 41.8 |
| 3 | GPT-5 (Low) | 35.8 |
| 4 | GPT-5 Mini (High) | 32.4 |
| 5 | GPT-5 Mini (Medium) | 31.8 |
| 6 | O4 Mini | 30.4 |
| 7 | GPT-5 Mini (Low) | 28.3 |
| 8 | Qwen 3 VL 32B (Thinking) | 22.2 |
| 9 | Qwen 3 VL 235B A22B (Thinking) | 21.1 |
| 10 | Qwen 3 VL 8B Instruct | 20.8 |
| 11 | Qwen 3 VL 30B A3B (Thinking) | 20.5 |
| 12 | GPT-5 Nano | 20.2 |
| 13 | GPT-4o | 18.6 |
| 14 | Qwen 3 VL 8B (Thinking) | 17.5 |
| 15 | Gemini 2.5 Flash | 16.3 |
Interactive version: theaggregate.ai/benchmark?slug=visreason-vision-centric-fill-in-the-blank · How It Works · Data refreshed daily, snapshot 2026-10-07.