iPrefBench Action Completion
Agents · 2026-10-26
iPrefBench's action-completion score: the fraction of each task's action-completion criteria satisfied, covering the operations the instruction requires regardless of user-specific choices (about two per task), macro-averaged over the 130 tasks, in percent. Same tasks, runs and Claude Sonnet 5 judge as task success.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Opus 5 | 91.6 |
| Kimi K3 | 85.8 |
| Sonnet 5 | 83.1 |
| Gemini 3.7 Flash | 79.6 |
| Gemini 3.1 Pro Preview | 78.6 |
| Qwen3.8-Flash-Next | 77.6 |
| Qwen3.8 27B | 77.5 |