OpenAI MentalHealthBench - Acuity - High Acuity: leaderboard
Acuity split of OpenAI's MentalHealthBench (launch post and paper, 2026-09-23): the 221 high-acuity conversations (serious concerns or significant distress, not an immediate emergency). Mean task-clipped rubric score, 0-100, higher is better: GPT-5.6 Sol grades each reply against expert-written signed rubrics, and models run at their API default reasoning effort. GPT-5 Thinking and Gemini 2.5 Flash come from the paper's Figure 7.
Metric: Task-clipped rubric score (%). Source: openai.com. Saturation forecast: Around 2029. 17 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-6 Astra | 57.92 | #5 |
| 2 | GPT-6 Sol | 57 | #17 |
| 3 | Muse Spark 1.3 | 55.5 | #31 |
| 4 | GPT-5.6 Sol (August 2026) | 53.59 | |
| 5 | GPT-6 Luna | 53.32 | #62 |
| 6 | Claude Opus 5.5 | 52.81 | #3 |
| 7 | GPT-5.6 Luna (August 2026) | 52.34 | |
| 8 | Claude Fable 5.1 | 50.53 | #10 |
| 9 | GPT-5 (Thinking) | 50.3 | #108 (GPT-5) |
| 10 | Claude Sonnet 5 | 48.66 | #67 |
| 11 | Grok 4.7 | 44.28 | #26 |
| 12 | Claude Haiku 4.5 | 43.71 | #291 |
| 13 | Gemini 3.8 Flash | 41.3 | #30 |
| 14 | Gemini 3.1 Pro (Preview) | 38.49 | #66 |
| 15 | Gemini 2.5 Flash | 37.7 | #255 |
No result here: #6 GPT-6.1 Sol, #9 Gemini 4 Argon.
Interactive version: theaggregate.ai/benchmark?slug=openai-mentalhealthbench-acuity-high-acuity · How It Works · Data refreshed daily, snapshot 2026-11-07.