MHGraphBench - Entity Clustering: leaderboard

Odd-one-out questions over five entities in MHGraphBench (2026), a knowledge-graph-grounded mental-health benchmark built from a curated PrimeKG subgraph, with letter-only multiple-choice answers. Accuracy in percent; higher is better.

Metric: Accuracy (%) on the 2,000 five-entity odd-one-out entity clustering items, letter-only answers; the OpenAI API models answer at temperature 0 (at most 120 completion tokens) with strict answer-letter parsing, the open models are scored by forced-choice option-letter log-probabilities; higher is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 15 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1GPT-5.1 Instant92.35#472
2GPT-4o91.85#751
3GPT-4.191.85#505
4GPT-5 Mini91.75#364
5GPT-5.2 Instant90.1#304
6Qwen 2.5 32B Instruct54.6#1247
7Qwen 2.5 7B Instruct36.95#2723
8Mistral 7B Instruct (v0.3)28.15#6750
9Llama3-Med42-8B24.55#5084
10Llama3-OpenBioLLM-8B23.35#6364
11DeepSeek-R1-Distill-Qwen-7B20.4#8055
12DeepSeek R1 Distill Qwen 32B18.9#1834
13Meditron-7B18#8733
14Llama 3.1 8B Instruct17.5#4600
15BioMistral-7B14.8#8332

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

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