MMOOC - Partial Answerability (Open-Ended VQA): leaderboard

Metric: Shifted in-context score (%; mean of accuracy on the answerable Partial Answerability Open-Ended VQA questions and answer rationality, the mean 0-1 judge score of the answer's correctness and evidence consistency in steps of 0.25; judge scores averaged over GPT-5.6, Claude Opus 5 and DeepSeek-V4-Pro). Source: arxiv.org. Saturation forecast: Around 2029. 18 models tracked.

Top models

#ModelScore
1Qwen 3.5 122B A10B72.25
2Gemma 4 31B70.75
3Gemma 4 26B67.5
4Ministral 3 8B64.25
5Qwen 3.5 27B63
6GPT-4o62.5
7Llama 4 Maverick58
8O156.5
9Ministral 3 14B55.5
10InternVL3-8B53
11O344.25
12Claude Opus 4.632.25
13Gemini 3.1 Pro (Preview)22.75

Interactive version: theaggregate.ai/benchmark?slug=mmooc-partial-answerability-open-ended-vqa · How It Works · Data refreshed daily, snapshot 2026-09-29.