MHGraphBench - Two-hop Verification: leaderboard

Two-hop verification questions, whether a drug relates to a disease through an intermediate disease, 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 verification items (about half yes), 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 2030. 15 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1GPT-4o62.08#754
2GPT-4.157.58#506
3Llama3-Med42-8B55#5086
4BioMistral-7B53.92#8342
5Qwen 2.5 32B Instruct50.5#1253
6Llama 3.1 8B Instruct50.42#4655
7GPT-5 Mini50.25#373
8GPT-5.2 Instant50.17#309
9GPT-5.1 Instant50.17#463
10Qwen 2.5 7B Instruct50.08#2732
11DeepSeek R1 Distill Qwen 32B49.75#1844
12DeepSeek-R1-Distill-Qwen-7B49.75#7976
13Llama3-OpenBioLLM-8B49.75#6375
14Mistral 7B Instruct (v0.3)49.33#6759
15Meditron-7B49.17#8747

No result here: #2 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=mhgraphbench-two-hop-verification · How It Works · Data refreshed daily, snapshot 2026-11-22.