K-Bench - Risk (Therapeutic v0): leaderboard
Metric: Risk Score (%, Therapeutic v0 prompt). Source: arxiv.org. Saturation forecast: Estimated already saturated. 63 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Fable 5 (Low) | 95.17 |
| 2 | GPT-5.5 (Non-reasoning) | 95.16 |
| 3 | GPT-5.2 (Non-reasoning) | 95.15 |
| 4 | GPT-5.4 (Low) | 94.38 |
| 5 | GPT-5.6 Terra (Non-reasoning) | 94.25 |
| 6 | Claude Fable 5 (High) | 94.15 |
| 7 | GPT-5.2 (High) | 94 |
| 8 | GPT-5.6 Luna (Non-reasoning) | 93.78 |
| 9 | GPT-5.5 (Low) | 93.72 |
| 10 | GPT-5.5 (Medium) | 93.64 |
| 11 | GPT-5.6 Sol (Non-reasoning) | 93.64 |
| 12 | GPT-5.4 (High) | 93.48 |
| 13 | GPT-5.6 Terra (High) | 93.47 |
| 14 | Claude Haiku 4.5 (Non-reasoning) | 93.41 |
| 15 | GPT-5.6 Terra (Medium) | 93.4 |
Interactive version: theaggregate.ai/benchmark?slug=k-bench-risk-therapeutic-v0 · How It Works · Data refreshed daily, snapshot 2026-10-04.