OpenAI MentalHealthBench - Behavior - Reality Testing: leaderboard
Behavior axis of OpenAI's MentalHealthBench (launch post and paper, 2026-09-23): not affirming implausible, paranoid or delusional beliefs as fact while acknowledging distress, scored on the 363 rubric criteria tagged with this axis. The post's axis score is the signed rubric score on this axis, normalized to its possible range (0-100, higher is better); GPT-5.6 Sol grades, models run at API default reasoning effort. The post's 15 models only: the paper's behavior figure uses a different scale.
Metric: Axis score, normalized to its possible range (%). Source: openai.com. Saturation forecast: Around February 2027. 15 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-6 Astra | 91.65 | #5 |
| 2 | GPT-6 Sol | 87.18 | #17 |
| 3 | GPT-6 Luna | 86.91 | #62 |
| 4 | GPT-5.6 Sol (August 2026) | 77.66 | |
| 5 | GPT-5.6 Luna (August 2026) | 76.85 | |
| 6 | Claude Opus 5.5 | 72.52 | #3 |
| 7 | Grok 4.7 | 71.37 | #26 |
| 8 | Claude Haiku 4.5 | 65.48 | #291 |
| 9 | Claude Sonnet 5 | 65.35 | #67 |
| 10 | Claude Fable 5.1 | 64.13 | #10 |
| 11 | Gemini 3.8 Flash | 59.36 | #30 |
| 12 | Muse Spark 1.3 | 54.95 | #31 |
| 13 | GPT-4o (Mar 2025) | 54.79 | #339 |
| 14 | Gemini 3.1 Pro (Preview) | 49.22 | #66 |
| 15 | Gemini 2.5 Pro | 42.16 | #161 |
No result here: #6 GPT-6.1 Sol, #9 Gemini 4 Argon.
Interactive version: theaggregate.ai/benchmark?slug=openai-mentalhealthbench-behavior-reality-testing · How It Works · Data refreshed daily, snapshot 2026-11-07.