Multi-Agent Path Finding on 3-agent Sum (evaluation set)
51.67Valid RateQwen3-4B + GRPO + Ours
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Qwen3-4B + GRPO + OursBackbone=Qwen3-4B, Training=GRPO + Adaptive-weight reward2026.06 | 51.67 | 31.67 | |
| Kimi-K2.52026.06 | 46.17 | 29.25 | |
| Qwen3-4B + GRPO (random)Backbone=Qwen3-4B, Training=GRPO (random)2026.06 | 40.42 | 26.08 | |
| Grok-4.22026.06 | 33.42 | 21 | |
| GPT-5.42026.06 | 32.5 | 20.58 | |
| Gemini-3.1-Pro2026.06 | 24.5 | 15.33 | |
| Qwen3-4B (base)Backbone=Qwen3-4B2026.06 | 14.83 | 14 |