Interactive Agent Task on TRIP-Bench Hard PMR
36Loose SuccessGPT-5.2
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| GPT-5.2thinking mode=w/ thinking2026.02 | 36 | 10 | |
| DeepSeek-V3.2thinking mode=w/ thinking2026.02 | 20 | 2 | |
| GLM-4.7thinking mode=w/ thinking2026.02 | 18 | 0 | |
| Claude-Sonnet-4.5thinking mode=w/ thinking2026.02 | 18 | 2 | |
| Gemini-3-Flashthinking mode=w/o thinking2026.02 | 14 | 0 | |
| Gemini-3-Flashthinking mode=w/ thinking2026.02 | 12 | 0 | |
| Gemini-3-Prothinking mode=w/o thinking2026.02 | 10 | 0 | |
| Gemini-3-Prothinking mode=w/ thinking2026.02 | 10 | 0 | |
| GPT-5.2thinking mode=w/o thinking2026.02 | 6 | 0 | |
| Claude-Sonnet-4.5thinking mode=w/o thinking2026.02 | 4 | 0 | |
| DeepSeek-V3.2thinking mode=w/o thinking2026.02 | 4 | 0 | |
| Kimi-K2-0905-Previewthinking mode=w/o thinking2026.02 | 0 | 0 | |
| Qwen3-235B-A22B-Instruct-2507thinking mode=w/o thinking2026.02 | 0 | 0 | |
| GLM-4.7thinking mode=w/o thinking2026.02 | 0 | 0 | |
| Qwen3-235B-A22B-Thinking-2507thinking mode=w/ thinking2026.02 | 0 | 0 | |
| Kimi-K2-Thinkingthinking mode=w/ thinking2026.02 | 0 | 0 |