HorizonBench - Evolved Preferences: leaderboard
Metric: Accuracy (%) on the 2,484 items whose target preference changed after a life event of HorizonBench (5-option questions on which assistant response fits the user's current preference, given six months of synthetic conversation history, about 163K tokens and 4,300 turns; pre-evolution values as hard distractors; items kept only if five validation models all failed them without the history; chance 20); higher is better. Source: arxiv.org. Saturation forecast: Not forecast. 25 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Opus 4.5 | 51.3 |
| 2 | Claude 3.7 Sonnet | 46.7 |
| 3 | Claude Opus 4 | 44.4 |
| 4 | Gemini 3.1 Pro (Preview) | 38.4 |
| 5 | Gemini 3 Pro | 38.4 |
| 6 | Claude Sonnet 4.5 | 32.6 |
| 7 | Claude 3.5 Haiku | 31.8 |
| 8 | Claude Sonnet 4 | 31.4 |
| 9 | Claude 3.5 Sonnet (20240620) | 29.2 |
| 10 | Gemini 2.0 Flash | 24.2 |
| 11 | GPT-4o | 23.9 |
| 12 | Gemini 2.5 Pro | 21.5 |
| 13 | Gemini 2.0 Flash Lite | 20.3 |
| 14 | Gemini 2.5 Flash Lite | 18.2 |
| 15 | O1 | 15.5 |
Interactive version: theaggregate.ai/benchmark?slug=horizonbench-evolved-preferences · How It Works · Data refreshed daily, snapshot 2026-10-07.