XTC-Bench - Understanding: leaderboard

XTC-Bench (2026) visual understanding on 2,000 COCO and Visual Genome images: one question per scene-graph fact (object retrieval, attributes, relations), answers judged 0-5 against the reference. Understanding score (0-1, shown times 100); higher is better.

Metric: Overall understanding score (0-1, times 100) on visual questions derived one per fact from the reference scene graph of the real image, a Qwen3-235B judge scores each fact 0-5, normalized to 0-1 and shown times 100, over XTC-Bench's 2,000 scene-graph-annotated images from COCO 2017 val and Visual Genome (over 31,000 atomic facts about objects, attributes and relations); higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 10 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Gemini 2.5 Flash Image74
2BAGEL-7B-MoT72.4#4619
3BLIP3-o8B69.2
4GPT-568.4#178
5Janus-Pro-7B68.1#6857
6OmniGen265.9#4579
7Show-o2-7B60.5
8Tar-7B58
9Show-o50.3
10MMaDA-8B38

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

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