GPTNT - Manual VQA Reading: leaderboard
Offline manual questions from GPTNT (Parekh et al., TMLR 2026): 191 questions that ask a model to read specific text or values from the Keep Talking and Nobody Explodes bomb manual. Accuracy in percent; higher is better.
Metric: Accuracy (%; 191 questions reading specific text or values from the bomb manual). Source: arxiv.org. Saturation forecast: Around February 2027. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Flash (Preview) (Minimal) | 81.2 | #141 (Gemini 3 Flash (Preview)) |
| 2 | Claude Sonnet 4.6 (Non-reasoning) | 75.9 | #158 (Claude Sonnet 4.6) |
| 3 | Qwen 3.5 27B (Non-reasoning) | 68.6 | #358 (Qwen 3.5 27B) |
| 4 | GPT-5.2 (2025-12-11) (Non-reasoning) | 65.4 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=gptnt-manual-vqa-reading · How It Works · Data refreshed daily, snapshot 2026-11-12.