Agent's Last Exampass@1

A hard set of desktop and operating-system tasks an AI agent has to finish by looking at the screen and working the machine itself. The score is the share it passes outright — partial credit does not count. Higher is better.

GPT-6 Astra by OpenAI holds the best Agent's Last Exam (pass@1) result among tracked models, at 59.3% (released Sep 3 2026). Higher scores are better on this benchmark.

1. GPT-6 Astra (OpenAI) — 59.3%; 2. DeepSeek-V4.1-Flash (DeepSeek) — 31.8%; 3. GLM-5.3 (Z.ai) — 28.5%; 4. Gemini 3.7 Flash (Google) — 26.3%; 4. GLM-5.3-Flash (Z.ai) — 26.3%.

7 tracked models have a published Agent's Last Exam (pass@1) score. Scores are the figures reported by each lab at that model's release, so this page is a record of results over time rather than a re-run leaderboard.

GLM-5.3 by Z.ai is the highest-ranked model with downloadable weights on Agent's Last Exam (pass@1), scoring 28.5% at rank 3 overall.