MHGraphBench - Two-hop Verification with Evidence: leaderboard
Two-hop verification 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 verification 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 April 2028. 15 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4.1 | 71.42 | #505 |
| 2 | GPT-4o | 68.83 | #751 |
| 3 | Qwen 2.5 32B Instruct | 61.25 | #1247 |
| 4 | GPT-5.2 Instant | 61.08 | #304 |
| 5 | GPT-5.1 Instant | 60 | #472 |
| 6 | GPT-5 Mini | 59.67 | #364 |
| 7 | Qwen 2.5 7B Instruct | 56.75 | #2723 |
| 8 | Llama3-OpenBioLLM-8B | 56.5 | #6364 |
| 9 | Mistral 7B Instruct (v0.3) | 56.17 | #6750 |
| 10 | BioMistral-7B | 53.17 | #8332 |
| 11 | Llama3-Med42-8B | 51.42 | #5084 |
| 12 | Meditron-7B | 51 | #8733 |
| 13 | Llama 3.1 8B Instruct | 50.25 | #4600 |
| 14 | DeepSeek-R1-Distill-Qwen-7B | 49.75 | #8055 |
| 15 | DeepSeek R1 Distill Qwen 32B | 49.58 | #1834 |
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-verification-with-evidence · How It Works · Data refreshed daily, snapshot 2026-11-21.