HyperGVL - Vertex Counting: leaderboard

HyperGVL (2026) hypergraph task for vision-language models: counting the vertices of a hypergraph, with each of 200 problems posed in 35 textual and visual representations. Accuracy (%); higher is better.

Metric: Accuracy (%) on counting the vertices of a hypergraph, 7,000 questions per task (200 problems on synthetic and real-world hypergraphs, each posed in 35 textual, visual and combined hypergraph representations), accuracy (%) averaged over the representations, temperature 0.8, top-p 0.95; higher is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 12 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Gemma 3 12B99.8#2059
2Gemini 3 Flash99.73#175
3Qwen 3 VL 8B99.6#917
4GPT-4o98.29#751
5Qwen 2.5 VL 7B Instruct95.41#1902
6GLM-4V-9B91.34#3142
7InternVL3-8B88.93#1782
8InternVL2.5-8B73.34#2370
9DeepSeek-VL248.59#2927
10Molmo-7B-D-092442.24#7207
11LLaVA-NeXT-7B19.56#5032
12LLaVA-v1.5-7B14.1#7630

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

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