Interactive LLM Agent Task Success on SciWorld Seen
86.4Success RateQ-Evolve
Evaluation results
| Method | Links | |
|---|---|---|
| Q-EvolveBackbone=Llama-3-8B-Instruct2026.06 | 86.4 | |
| ETO + MPOBackbone=Llama-3-8B-Instruct2026.06 | 83.4 | |
| WKMBackbone=Llama-3-8B-Instruct2026.06 | 82.1 | |
| KnowAgentBackbone=Llama-3-8B-Instruct2026.06 | 81.7 | |
| ETOBackbone=Llama-3-8B-Instruct2026.06 | 81.3 | |
| SFT + MPOBackbone=Llama-3-8B-Instruct2026.06 | 70.2 | |
| SFTBackbone=Llama-3-8B-Instruct2026.06 | 65.3 |