DailyClue - Scientific Commonsense: leaderboard

Visual clue-driven reasoning on DailyClue (2026, Seek-and-Solve), scientific commonsense reasoning subset: everyday images whose question can only be answered after finding a decisive visual clue, with items dropped when all three filter models answered them correctly. Accuracy in percent; higher is better.

Metric: Accuracy (%) on scientific commonsense reasoning (physics, chemistry and biology in everyday scenes) on DailyClue, 666 question-image pairs from daily scenarios whose answer requires finding a decisive visual clue; question-clue-answer triplets drafted by GPT-5 and Gemini 2.5 Pro, manually verified, and kept only when at most two of o4-mini, Gemini 2.5 Flash and Claude 3.7 Sonnet answered correctly; exact match, with a Gemini 2.5 Pro judge for open-ended answers; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 24 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Gemini 2.5 Pro67.48#276
2GPT-561.79#178
3Gemini 2.5 Flash56.91#463
4Qwen 3 VL 235B A22B (Thinking)56.1#459
5O4 Mini50.41#368
6Claude Sonnet 449.59#410
7Qwen 3 VL 235B A22B Instruct48.78#532
8Claude Sonnet 4.547.97#261
9Claude 3.7 Sonnet47.97#470
10Qwen 2.5 VL 72B Instruct47.15#841
11InternVL3.5-38B43.9#1021
12InternVL3-78B42.28#759
13LLaVA-OneVision-72B42.28#1771
14InternVL3-8B41.46#1782
15InternVL3-38B39.84#939

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=dailyclue-scientific-commonsense · How It Works · Data refreshed daily, snapshot 2026-11-21.