RL-trained code search agents (1.7B, 4B, 14B) that outperform 2β18Γ larger models using only a Unix terminal. π arxiv.org/abs/2603.17829
-
OpenHands/CodeScout-14B
Text Generation β’ 15B β’ Updated β’ 70 β’ β’ 3 -
OpenHands/CodeScout-4B
Text Generation β’ 4B β’ Updated β’ 23 β’ β’ 1 -
OpenHands/CodeScout-1.7B
Text Generation β’ 2B β’ Updated β’ 11 β’ β’ 1 -
OpenHands/CodeScout-1.7B-RFT
Text Generation β’ 2B β’ Updated β’ 14 β’ β’ 1