PM-Bench (Auto Heartbeat 60m): leaderboard

Metric: Set-F1 (%; F1 of the prospective actions the agent selects against the actions actually due, accumulated over the 80 steps and 81 scored tasks of the released simulated week, lures included; single agent with a heartbeat switched on every 60 virtual minutes; one run per model). Source: arxiv.org. Saturation forecast: Around December 2026. 8 models tracked.

Top models

#ModelScore
1GPT-5.3 Codex74.8
2Mistral-Large-3-675B-Instruct-251269.3
3GPT-5.463.8
4Llama 3.3 70B Instruct60.7
5Qwen 3 32B55.2
6Mistral Small 3.252.7
7Qwen 3 8B42.5
8Qwen 3 14B34

Interactive version: theaggregate.ai/benchmark?slug=pm-bench-auto-heartbeat-60m · How It Works · Data refreshed daily, snapshot 2026-09-29.