RL-trained code search agents (1.7B, 4B, 14B) that outperform 2ā18Ć larger models using only a Unix terminal. š arxiv.org/abs/2603.17829
-
OpenHands/CodeScout-14B
Text Generation ⢠15B ⢠Updated ⢠36 ⢠⢠3 -
OpenHands/CodeScout-4B
Text Generation ⢠4B ⢠Updated ⢠11 ⢠⢠1 -
OpenHands/CodeScout-1.7B
Text Generation ⢠2B ⢠Updated ⢠13 ⢠⢠1 -
OpenHands/CodeScout-1.7B-RFT
Text Generation ⢠2B ⢠Updated ⢠16 ⢠⢠1