Blind-Spots-Bench - Multimodal-to-Text: leaderboard
Metric: Accuracy (%; mean@4 over four samples on the questions with image input and text output, vision-language models only; 235 human-authored questions with reference solutions graded correct or incorrect by gemini-3-flash with code execution; no tools; thinking enabled at medium effort where available; max 32,768 output tokens). Source: arxiv.org. Saturation forecast: Around December 2026. 22 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) (Medium) | 66.9 |
| 2 | Gemini 3 Flash (Preview) (Medium) | 64.5 |
| 3 | GPT-5.5 | 58.7 |
| 4 | GPT-5.4 (Medium) | 58.1 |
| 5 | GPT-5.2 (Medium) | 50 |
| 6 | Qwen 3.5 122B A10B | 48.8 |
| 7 | Qwen 3.5 397B A17B | 48.3 |
| 8 | Qwen 3.5 35B A3B | 47.1 |
| 9 | GPT-5.4 Mini (Medium) | 45.3 |
| 10 | Kimi K2.5 | 42.4 |
| 11 | Kimi K2.6 | 41.9 |
| 12 | gemma-4-26B-A4B-it (Thinking) | 40.7 |
| 13 | GPT-5 | 37.8 |
| 14 | Gemini 2.5 Pro | 36 |
| 15 | GPT-5 Mini | 33.1 |
Interactive version: theaggregate.ai/benchmark?slug=blind-spots-bench-multimodal-to-text · How It Works · Data refreshed daily, snapshot 2026-09-29.