PAST-Bench: leaderboard
Metric: Overall task score (%; mean of the four capability scores: memory, procedural reuse, information gathering and update; persistence on: the agent may reuse memory records, skills, profile state and session history produced by earlier episodes of the family; 26 scenarios and 204 episodes run as ordered sequences of fresh-session tasks; each evaluation episode scores completion (audit-data checks, or a MiniMax-M2.7 rubric judge for open-ended families), tool-error recovery and a safety gate; family-balanced means over three trials; Hermes agent framework). Source: arxiv.org. Saturation forecast: Around December 2026. 7 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5.4 (Medium) | 75 |
| 2 | Kimi K2.6 (Thinking) | 73 |
| 3 | Claude Sonnet 4.6 | 71 |
| 4 | GLM-5.1 | 71 |
| 5 | DeepSeek V4 Pro (High) | 67 |
| 6 | Claude Opus 4.6 (Medium) | 67 |
| 7 | MiniMax-M2.7 | 66 |
Interactive version: theaggregate.ai/benchmark?slug=past-bench · How It Works · Data refreshed daily, snapshot 2026-09-29.