Agents' Last Exam Near-Term Pass Rate
Agents · 2026-06-03
Perfect-score pass rate on the 67-task near-term subset containing workflows that frontier agents can partially complete today.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Opus 5.5 | 53.7 |
| GPT-6 Astra | 52.2 |
| GPT-6 Sol | 50.7 |
| Claude Opus 5 | 49.3 |
| GPT-5.6 Luna | 49.3 |
| GPT-5.6 Sol | 47.8 |
| Muse Spark 1.3 | 47.8 |
| MiniMax M3.1 Flash | 46.3 |
| GPT-5.6 Terra | 44.8 |
| GPT-5.5 | 44.0 |
| Opus 4.8 | 43.3 |
| Kimi K3 | 40.3 |
| GPT-6 Luna | 38.8 |
| Claude Fable 5 | 37.3 |
| Composer 2.5 | 34.3 |
| GPT-5.4 | 33.6 |
| GLM-5.2 | 32.8 |
| Agnes 2.5 Pro Beta | 31.3 |
| Opus 4.7 | 31.3 |
| Gemini 3.1 Pro Preview | 28.4 |
| Doubao Seed 2.1 Pro | 26.9 |
| GLM-5.1 | 20.1 |
| DeepSeek-V4-Pro | 19.9 |
| Qwen3.7-Max (source unspecified) | 17.9 |
| Kimi K2.6 | 15.7 |
| Qwen3.6 Plus (2026-04-02) | 12.7 |
| MiMo-V2.5 | 11.9 |
| Grok 4.3 | 10.4 |
| MiniMax M2.7 | 10.4 |