Agent Behavior Adaptation on WebShop (WS) (test)
36.7Loop RatioQwen3-4B-Instruct
Evaluation Results
| Method | Links | |
|---|---|---|
| Qwen3-4B-InstructModel Type=Non-thinking Model2026.02 | 36.7 | |
| Phi-4-reasoningModel Type=Thinking Model2026.02 | 35.7 | |
| Mistral-7B-InstructModel Type=Non-thinking Model2026.02 | 17.5 | |
| gpt-oss-120bModel Type=Thinking Model2026.02 | 9.2 | |
| Qwen3-30B-A3B-InstructModel Type=Non-thinking Model2026.02 | 5.7 | |
| Qwen3-4B-ThinkingModel Type=Thinking Model2026.02 | 4.5 | |
| Ministral-3-14B-InstructModel Type=Non-thinking Model2026.02 | 4 | |
| Phi-4Model Type=Non-thinking Model2026.02 | 2.6 | |
| Llama-3.1-8B-InstructModel Type=Non-thinking Model2026.02 | 2.1 | |
| Llama-3.3-70B-InstructModel Type=Non-thinking Model2026.02 | 1.6 | |
| Glm-4-9B-ChatModel Type=Non-thinking Model2026.02 | 1.5 | |
| Qwen3-30B-A3B-ThinkingModel Type=Thinking Model2026.02 | 0.7 | |
| Gemini 2.5 FlashModel Type=Non-thinking Model2026.02 | 0.2 | |
| Glm-4-32B-0414Model Type=Non-thinking Model2026.02 | 0.1 | |
| DeepSeek-R1Model Type=Thinking Model2026.02 | 0.1 | |
| DeepSeek-V3.2Model Type=Non-thinking Model2026.02 | 0 | |
| Gemini 2.5 ProModel Type=Thinking Model2026.02 | 0 |