Agents' Last Exam (ALE)
Agents · 2026-06-03
Agents' Last Exam evaluates generalist computer-use agents on 152 long-horizon, economically valuable professional workflows with hidden references and verifiable outcome-based grading. This headline metric is the pass rate on the full task set: the share of tasks receiving a perfect rubric score.
Top models (higher is better)
| Model | Score |
|---|---|
| Gemini 4 Argon | 39.5 |
| Claude Opus 5.5 | 38.2 |
| GPT-6 Astra | 34.2 |
| Claude Opus 5 | 32.2 |
| GPT-6 Sol | 32.2 |
| Muse Spark 1.3 | 32.2 |
| GPT-5.6 Sol | 30.6 |
| GPT-5.6 Luna | 30.3 |
| MiniMax M3.1 Flash | 29.6 |
| Kimi K3 | 28.3 |
| GPT-5.6 Terra | 28.0 |
| Opus 4.8 | 27.0 |
| GPT-5.5 | 26.6 |
| Claude Fable 5 | 25.7 |
| GPT-6 Luna | 25.0 |
| Agnes 2.5 Pro Beta | 21.7 |
| Opus 4.7 | 21.1 |
| GPT-5.4 | 20.5 |
| Composer 2.5 | 20.4 |
| GLM-5.2 | 20.4 |
| Doubao Seed 2.1 Pro | 19.5 |
| Gemini 3.1 Pro Preview | 16.4 |
| DeepSeek-V4-Pro | 12.4 |
| Qwen3.7-Max (source unspecified) | 11.8 |
| GLM-5.1 | 11.5 |
| Kimi K2.6 | 9.2 |
| MiMo-V2.5 | 8.6 |
| Qwen3.6 Plus (2026-04-02) | 8.6 |
| Grok 4.3 | 7.2 |
| MiniMax M2.7 | 5.9 |