Agents' Last Exam v1.1 (ALE v1.1)
Agents
Agents' Last Exam v1.1 is the maintainers' 122-task release of the long-horizon professional-workflow benchmark for generalist computer-use agents, replacing the 152-task v1.0 board (now at /leaderboard/v1). This headline metric is the pass rate: the share of runs earning a perfect rubric score, averaged over every release task, with a failed run scoring zero; each run has a hard five-hour wall-clock cap.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Opus 5.5 | 42.6 |
| GPT-6 Astra | 39.3 |
| Gemini 4 Argon | 37.7 |
| Muse Wonder | 36.9 |
| Kimi K3 | 35.2 |
| GPT-5.6 Sol | 30.3 |
| MiniMax M3.1 Flash | 29.5 |
| GLM-5.3 | 26.2 |
| GPT-5.6 Luna | 5.7 |