Agentic Coding on SWE-Bench Lite
24Solve RateCONTEXTRL
Evaluation results
| Method | Links | |
|---|---|---|
| CONTEXTRLTraining Source=Klear-AgentForge-8B, Training Strategy=CONTEXTRL2026.06 | 24 | |
| Qwen3-Coder-30BTraining Source=Off-the-shelf reference models2026.06 | 22 | |
| RL baselineTraining Source=Klear-AgentForge-8B, Training Strategy=RL baseline2026.06 | 21.7 | |
| Base modelTraining Source=Klear-AgentForge-8B, Training Strategy=Base model2026.06 | 21 | |
| Qwen3-14BTraining Source=Off-the-shelf reference models2026.06 | 6 | |
| Qwen3-32BTraining Source=Off-the-shelf reference models2026.06 | 6 | |
| CONTEXTRLTraining Source=Qwen3-8B, Training Strategy=CONTEXTRL2026.06 | 4 | |
| Base modelTraining Source=Qwen3-8B, Training Strategy=Base model2026.06 | 2.7 | |
| RL baselineTraining Source=Qwen3-8B, Training Strategy=RL baseline2026.06 | 2.7 |