PM-Bench (Todo Ledger): 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 that carries and rewrites an in-context ledger of pending intentions every step; one run per model). Source: arxiv.org. Saturation forecast: Around December 2026. 8 models tracked.

Top models

#ModelScore
1GPT-5.3 Codex74.8
2GPT-5.473.5
3Llama 3.3 70B Instruct72.5
4Mistral-Large-3-675B-Instruct-251266.7
5Mistral Small 3.263.4
6Qwen 3 14B52.3
7Qwen 3 32B51.7
8Qwen 3 8B47.8

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