MLExpert: 35B Agentic Model for Autonomous Machine Learning Engineering
MLExpert: Towards a Generalist Agentic Model for Autonomous Machine Learning Engineering
A 35B model matching frontier-scale systems on MLE-Bench Lite suggests agentic MLE performance depends more on training method than raw parameter count, lowering the hardware bar for teams building autonomous experimentation agents. The WIRL reward design over multi-turn execution windows is a reusable recipe for anyone training long-horizon agents beyond ML engineering. The paper also ships an executable task factory, so practitioners can adapt the harness to their own domains.
A new paper introduces MLExpert, a 35B Mixture-of-Experts agentic model trained for autonomous long-horizon machine learning engineering. The training pipeline combines the Kaggle-MLE Factory for executable task construction, hierarchical agentic SFT, and Windowed Interaction RL (WIRL), which assigns shared outcome-based rewards across multi-turn execution windows. On MLE-Bench Lite, MLExpert raises the base model's average medal rate from 37.9% to 72.7%, matching larger frontier models such as Kimi-K3 and GLM-5.3. The learned optimization capability also transfers to NatureBench Lite, agent scaffold optimization and algorithm optimization.
MLExpert: Towards a Generalist Agentic Model for Autonomous Machine Learning Engineering
Large language models (LLMs) are expanding from assisting scientific research and engineering to autonomously developing and improving AI systems, advancing AI for AI (AI4AI) as a pathway toward recursive self-improvement (RSI). Workflow-driven machine learning engineering (MLE) systems provide a practical path toward this pursuit through complex orchestration that manages long-horizon experimentation, while recent MLE systems adopt model-directed interaction, allowing models to autonomously organize experiments and adapt optimization strategies based on execution feedback. However, feedback from each experiment reflects several preceding decisions, and a long MLE run contains many such costly experiments. Training must therefore select useful supervision from long interaction trajectories and learn from intermediate outcomes. We propose a post-training method for autonomous long-horizon MLE optimization and apply it to train MLExpert, a 35B Mixture-of-Experts agentic model. We design the Kaggle-MLE Factory to construct executable tasks and an interaction harness for model-directed experimentation. Hierarchical agentic SFT transfers general tool-use and coding capabilities to MLE through outcome-guided supervision. Windowed Interaction RL (WIRL) treats the multi-turn decisions between successive experiment results as an execution window and assigns it a shared outcome-based reward. Asynchronous collection and policy correction support policy updates before the trajectory ends. On MLE-Bench Lite, MLExpert improves the average medal rate of the base model from 37.9% to 72.7%, achieving performance comparable to larger frontier models such as Kimi-K3 and GLM-5.3. Further experiments show that MLExpert improves performance on NatureBench Lite, agent scaffold optimization, and algorithm optimization, suggesting that the learned optimization capability generalizes to other tasks.
