Agents' Last Exam — leaderboard
Long-horizon agent benchmark covering economically valuable professional workflows in full operating-system sandboxes, with verifiable task outcomes.
Metric: Pass Rate (%). Source: snorkel.ai. Status: saturation imminent. 20 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5.6 Sol | 30.6 |
| 2 | Claude Opus 4.8 | 27 |
| 3 | GPT-5.5 | 26.6 |
| 4 | Claude Fable 5 | 25.7 |
| 5 | GPT-5.4 | 20.5 |
| 6 | Claude Opus 4.7 | 20.4 |
| 7 | GLM-5.2 | 20.4 |
| 8 | Seed 2.1 Pro | 19.5 |
| 9 | Gemini 3.1 Pro (Preview) | 15.8 |
| 10 | DeepSeek V4 Pro | 12.4 |
| 11 | Qwen 3.7 Max | 11.8 |
| 12 | GLM-5.1 | 11.5 |
| 13 | Kimi K2.6 | 9.2 |
| 14 | Qwen 3.6 Plus | 8.6 |
| 15 | MiMo-V2.5 | 8.6 |
Interactive version: theaggregate.ai/benchmark?slug=agents-last-exam · How the rankings work · Data refreshed daily, snapshot 2026-07-22.