MHGraphBench - Two-hop Selection with Evidence: leaderboard

Two-hop selection questions with sanitized PrimeKG feature snippets added, 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 1,200 two-hop selection items with sanitized PrimeKG feature snippets added, 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: Around February 2027. 15 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1GPT-5.2 Instant67.58#304
2GPT-5 Mini65.42#364
3GPT-5.1 Instant64.42#472
4GPT-4.161.5#505
5GPT-4o61.42#751
6Qwen 2.5 32B Instruct50.08#1247
7BioMistral-7B38.42#8332
8Qwen 2.5 7B Instruct37.75#2723
9Llama3-Med42-8B36.17#5084
10DeepSeek R1 Distill Qwen 32B35.75#1834
11Llama 3.1 8B Instruct31.08#4600
12Mistral 7B Instruct (v0.3)28.25#6750
13DeepSeek-R1-Distill-Qwen-7B25#8055
14Llama3-OpenBioLLM-8B24.33#6364
15Meditron-7B22.08#8733

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

Interactive version: theaggregate.ai/benchmark?slug=mhgraphbench-two-hop-selection-with-evidence · How It Works · Data refreshed daily, snapshot 2026-11-21.