ReTerm: Recovery Training for Self-Improving Terminal Agents

Open paper
Benchmark Model Rank Results
mathematical-reasoning-on-aime-2025ReTerm-9B#13Accuracy: 93.3
mathematical-reasoning-on-aime24ReTerm-9B#19Acc: 86.7
code-generation-on-swe-bench-verifiedReTerm-9B (ReAct)#67Accuracy: 58.4

Rank counts only results with a code link.