LEGO-RL: Harness-Native Reinforcement Learning for Coding Agents
Open paper
Benchmark
Model
Rank
Results
code-generation-on-swe-bench-verified
Lego-RL-Qwen3.5-35B-A3B (OpenHands SDK) (extra data)
#42
Accuracy: 70.4
Rank counts only results with a code link.