GPTNT - Manual VQA Reading: leaderboard
Metric: Accuracy (%; 191 questions reading specific text or values from the bomb manual). Source: arxiv.org. Saturation forecast: Around December 2026. 5 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3 Flash (Preview) (Minimal) | 81.2 |
| 2 | Claude Sonnet 4.6 (Non-reasoning) | 75.9 |
| 3 | Qwen 3.5 27B (Non-reasoning) | 68.6 |
| 4 | GPT-5.2 (Non-reasoning) | 65.4 |
Interactive version: theaggregate.ai/benchmark?slug=gptnt-manual-vqa-reading · How It Works · Data refreshed daily, snapshot 2026-09-26.