Agent Workflow Execution on CUA-World-Long (test)
55.5Average ScoreGPT-5.4
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| GPT-5.4Max steps=2000, Cost cap=none2026.04 | 55.5 | 27.5 | |
| Gemini 3 FlashMax steps=2000, Cost cap=none2026.04 | 38.7 | 11.5 | |
| Gemini 3 FlashMax steps=500, Cost cap=$52026.04 | 36.2 | 7.5 | |
| Kimi-K 2.5Max steps=500, Cost cap=$52026.04 | 33.9 | 5.5 | |
| GPT-5.4Max steps=500, Cost cap=$52026.04 | 22.7 | 3 | |
| Sonnet 4.6Max steps=500, Cost cap=$52026.04 | 20.5 | 6 |