Frontis-MA1: Training an AI4AI Model towards Recursive Self-Improvement in Machine Learning Engineering

Abstract | Recursive self-improvement (RSI) requires AI systems that improve the process of building AI (i.e., AI4AI); machine learning engineering (MLE) offers a concrete, executable testbed for studying this capability. We introduce OpenMLE, an open full-stack system for RSI research in MLE, spanning verifiable task environments with execution feedback (OpenMLE-Gym), operator learning (OpenMLE-ERL), and long-horizon search (OpenMLE-Evo). On this stack we posttrain Frontis-MA1-35B as a meta-evolution agent for MLE, aligning post-training and inference around four atomic program-evolution operators (Draft, Improve, Debug, Crossover): the same operators are trained via execution-grounded SFT and RL on data deduplicated against all evaluation benchmarks, then composed into long-horizon search, coupling learning and evolution in a single loop. On MLE-Bench Lite under a 12-hour per-task budget on one RTX 4090 capped at 12 GB VRAM, Frontis-MA1-35B improves Medal Average from 39.39% to 60.61% over its base model with OpenMLE-Evo, and reaches 71.21% with OpenMLE-Evo-Max (benchmark-independent experience priors and asynchronous search), exceeding GPT-5.5 + Codex and approaching GPT-5.6 Sol and the 2.8T Kimi K3.

Introduction. AI capability growth is no longer pushed only by human engineers. Increasingly, AI systems write code, run experiments, search over designs, and help build the next generation of AI systems [Lu et al., 2024, Novikov et al., 2025, Oh et al., 2025, Romera-Paredes et al., 2024]. This broader direction, often called AI for AI (AI4AI), seeks to use AI systems to build and improve AI [Chan et al., 2026, Liu et al., 2025a]. Its more ambitious endpoint is recursive self-improvement (RSI), where each improved system further improves the process that produces its successors [Eth and Davidson, 2025, Favaro and Clark, 2026, Good, 1965, Schmidhuber, 2003]. Reaching that endpoint requires more than stronger one-shot generation or planning. It requires agents that can perform AI training AI and AutoResearch: inspect data, propose algorithms, execute experiments, diagnose failures, and decide how to spend the next unit of compute [Karpathy, 2026, Lu et al., 2024, Nathani et al., 2025].

Discussion / Conclusion. We presented OpenMLE, an open full-stack technical solution for training and deploying languagemodel agents that construct and iteratively improve machine learning solutions through executable feedback. OpenMLE-Gym provides quality-gated tasks, isolated execution, and task-specific evaluation; OpenMLE-ERL learns Draft, Improve, Debug, and Crossover transformations through execution-grounded supervised fine-tuning and reinforcement learning; and OpenMLE- Evo composes the same operators into long-horizon search using structured experience, multifactor parent selection, and operator-conditioned memory. This shared operator and execution interface makes Frontis-MA1-35B both the product of the training stack and the variation engine of its evolutionary harness. The results show that model learning and search provide complementary gains.