PersonaTrail (ReasoningBank) - Multi-hop Preference Inference: leaderboard
Metric: Task success rate (%; multi-hop preference inference over several websites; success needs every website, the intent and the preference satisfied; live websites through a Browser-Use web agent with at most 30 steps and temperature 0, given memory from ReasoningBank (per-user reasoning items distilled from successful and failed trajectories); Gemini 3.1 Pro evaluator; the three proprietary backbones run on a 25% stratified subset). Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 6 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Browser Use + Qwen3.6-27B | 13.73 |
| 2 | Browser Use + Claude Haiku 4.5 | 9.44 |
| 3 | Browser Use + Gemini-3-Flash | 9.19 |
| 4 | Browser Use + GPT-5.4-mini | 1.91 |
| 5 | Browser Use + Gemma-3-27B-it | 0.39 |
| 6 | Browser Use + Llama-4-Scout | 0.39 |
Interactive version: theaggregate.ai/benchmark?slug=personatrail-reasoningbank-multi-hop-preference-inference · How It Works · Data refreshed daily, snapshot 2026-09-29.