See-or-Guess VQA - Question-Guided: leaderboard
Metric: Question-Guided variant (Gemini-2.5-Pro rewrite of the source question into a harder, more image-dependent form): accuracy (%) with the image over 540 images from 21 public benchmarks in six reasoning categories; Qwen3-14B judges answer equivalence (yes/no), lenient extraction of answers after a missing </think> tag. Source: arxiv.org. Saturation forecast: Around October 2027. 11 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Qwen 3.5 397B A17B | 48.4 |
| 2 | GPT-5 Mini | 46.3 |
| 3 | Gemini 2.5 Pro | 46.1 |
| 4 | Qwen 3.5 9B | 45.7 |
| 5 | Claude Sonnet 4.6 | 41.7 |
| 6 | Gemini 3.1 Flash Lite | 37 |
| 7 | InternVL3.5-8B | 20.6 |
| 8 | Phi-4 Multimodal Instruct | 19.1 |
Interactive version: theaggregate.ai/benchmark?slug=see-or-guess-vqa-question-guided · How It Works · Data refreshed daily, snapshot 2026-09-29.