Single Product shopping assistance on SHOPPING COMPANION benchmark
90AccuracyQwen3-4B-LoRA + RL (Dual&Tool-wise Reward)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Qwen3-4B-LoRA + RL (Dual&Tool-wise Reward)Category=Ours2026.03 | 90 | 84 | |
| Qwen3-4B-LoRA + RL (Dual-reward)Category=Ours2026.03 | 89 | 81 | |
| GPT-4.1Category=Closed2026.03 | 88 | 78 | |
| GPT-5Category=Closed2026.03 | 82 | 75 | |
| Qwen3-4B-LoRACategory=Ours2026.03 | 82 | 72 | |
| Qwen3-MaxCategory=Closed2026.03 | 80 | 72 | |
| GPT-4oCategory=Closed2026.03 | 79 | 72 | |
| Qwen3-Next-80B-A3BCategory=Open2026.03 | 63 | 57 | |
| Qwen3-30B-A3BCategory=Open2026.03 | 60 | 53 | |
| Qwen3-4BCategory=Open2026.03 | 49 | 44 |