K-Bench (Therapeutic v0): leaderboard
Metric: Overall (%, Therapeutic v0 prompt). Source: arxiv.org. Saturation forecast: Estimated already saturated. 63 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5.2 (Non-reasoning) | 98.95 |
| 2 | GPT-5.5 (Low) | 98.72 |
| 3 | GPT-5.5 (Medium) | 98.65 |
| 4 | GPT-5.2 (High) | 98.64 |
| 5 | GPT-5.4 (Low) | 98.51 |
| 6 | Kimi K2.5 | 98.45 |
| 7 | Claude Haiku 4.5 (Non-reasoning) | 98.45 |
| 8 | GPT-5.5 (Non-reasoning) | 98.41 |
| 9 | GPT-5.4 (High) | 98.39 |
| 10 | GPT-5.5 (High) | 98.37 |
| 11 | GPT-5.4 (Non-reasoning) | 98.37 |
| 12 | Kimi K2.6 | 98.11 |
| 13 | Claude Fable 5 (Low) | 98.11 |
| 14 | GPT-5.6 Terra (Non-reasoning) | 98.05 |
| 15 | Claude Opus 4.8 (Low) | 97.87 |
Interactive version: theaggregate.ai/benchmark?slug=k-bench-therapeutic-v0 · How It Works · Data refreshed daily, snapshot 2026-10-04.