Microsoft trained a 4B coding agent almost entirely with Reinforcement Learning, without a bigger teacher

Microsoft Research Montréal, in collaboration with Mila and UC San Diego, developed FrogNano, a 4-billion-parameter coding agent. This model, based on Qwen3.5-4B, was trained almost entirely using reinforcement learning without human-labeled data or a larger "teacher" model. Its learning involved synthetic software engineering tasks generated and refined across approximately 1,500 environments through repeated rounds of task creation and RL, as detailed in arXiv:2609.07925 [cs.AI].