The starting point is the open Qwen3-32B model. Instead of SFT, reinforcement learning is used (the rLLM framework): the agent solves coding tasks in an execution environment, and the reward signal comes from verifiable correctness (e.g. passing tests / solving a SWE-Bench task). The model learns to plan, edit code and verify results over many steps. Test-time scaling is additionally used, e.g. generating and selecting among multiple solutions, which raises success (Pass@16, 59% with test-time scaling).
Strong coding agents usually require costly supervised fine-tuning on large demonstration datasets. DeepSWE shows they can be obtained purely via RL from an open model, fully openly and reproducibly.
The open language model from which agent training starts.
Official
Pure reinforcement learning in the rLLM framework, without supervised fine-tuning.
A reward signal from objective verification of solutions (passing tests).
Generating and selecting among multiple solutions at inference for higher success.
Official
RL needs a reliable, objective signal (e.g. tests); for tasks without clear verification the approach is harder.
Training an RL agent in execution environments is computationally and engineering-wise expensive.
On July 2, 2025 an open coding agent trained purely with RL from Qwen3-32B was released, with a full training recipe and weights; SOTA among open-weight agents on SWE-Bench-Verified.
