MentalBench (DSM Diagnosis) - Resolvable Differential Diagnosis: leaderboard
Metric: Accuracy (%) on the 13,050 Type 4 cases (cases that include the discriminating evidence, so only one diagnosis of the pair is correct; choose one or more); cases synthesized from a psychiatrist-built DSM-5 knowledge graph by three LLMs and validated by a psychiatrist and a clinical psychologist; four options per case; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 22 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 Mini | 77.72 | #176 |
| 2 | Claude Sonnet 4.5 | 74.84 | #138 |
| 3 | GPT-5.1 | 70.3 | #131 |
| 4 | Claude Haiku 4.5 | 65 | #271 |
| 5 | Gemini 2.5 Flash | 60.96 | #237 |
| 6 | Gemini 2.5 Pro | 59.82 | #145 |
| 7 | GPT-4o | 48.3 | #333 |
| 8 | Llama 3.1 70B Instruct | 34.97 | #548 |
| 9 | Qwen 2.5 14B Instruct | 31.83 | #634 |
| 10 | Qwen 2.5 32B Instruct | 30.89 | #491 |
| 11 | Qwen 3 235B A22B | 28.63 | #304 |
| 12 | Qwen 3 32B | 27.37 | #424 |
| 13 | Gemma 3 4B (IT) | 18.41 | #971 |
| 14 | Qwen 2.5 72B Instruct | 18.25 | #436 |
| 15 | Gemma 3 12B (IT) | 15.79 | #655 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=mentalbench-dsm-diagnosis-resolvable-differential-diagnosis · How It Works · Data refreshed daily, snapshot 2026-10-11.