Proactive Agent Task Execution on Pare-Bench 1.0 (full)
64.3Success@4Gemini 3 Flash
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Gemini 3 Flash2026.04 | 64.3 | 16.1 | 42.1 | 19.1 | 67.1 | 21.7 | |
| Claude 4.5 Sonnet2026.04 | 60.8 | 18.2 | 42 | 12.8 | 78.2 | 20.2 | |
| Gemini 3 Pro2026.04 | 59.4 | 11.9 | 35.1 | 16.5 | 71.2 | 19.5 | |
| GPT-52026.04 | 57.3 | 17.5 | 37.4 | 28.1 | 70.2 | 20.6 | |
| Qwen 3 4B Instruct2026.04 | 35 | 6.3 | 18.5 | 20.5 | 63.7 | 16.7 | |
| Llama 3.2 3B Instruct2026.04 | 23.8 | 1.4 | 10 | 23 | 58.4 | 16.6 | |
| Gemma 3 4B Instruct2026.04 | 7.7 | 0.7 | 3 | 14.2 | 17.6 | 8.8 |