DocVQA 2026 - Science Poster: leaderboard

DocVQA 2026 accuracy on science poster documents under the published high-thinking baseline protocol.

Metric: Accuracy (%, high thinking). Source: huggingface.co. Saturation forecast: Around November 2027. 4 models tracked.

Top models

#ModelScoreOverall rank
1Gemini 3 Pro (Preview) (High)30#71 (Gemini 3 Pro (Preview))
2Gemini 3 Flash (Preview) (High)20#90 (Gemini 3 Flash (Preview))
3GPT-5.2 (High)0#115 (GPT-5.2)
4GPT-5 Mini (High)0#193 (GPT-5 Mini)

No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.

Interactive version: theaggregate.ai/benchmark?slug=docvqa-2026-science-poster · How It Works · Data refreshed daily, snapshot 2026-11-05.