Frontis-MA1：训练面向机器学习工程递归自我改进的AI4AI模型

摘要

递归自我改进（RSI）需要能够改进AI构建过程的AI系统（即AI4AI）；机器学习工程（MLE）为研究这一能力提供了具体且可执行的测试平台。我们提出了OpenMLE，一个用于MLE中RSI研究的开源全栈系统，涵盖带执行反馈的可验证任务环境（OpenMLE-Gym）、算子学习（OpenMLE-RL）和长时程搜索（OpenMLE-Evo）。在这一技术栈上，我们将Frontis-MA1（35B）后训练为面向MLE的元演化智能体，使后训练和推理围绕四个原子程序演化算子进行对齐——起草（Draft）、改进（Improve）、调试（Debug）、交叉（Crossover）：这些算子通过基于执行的监督微调（SFT）和强化学习（RL），在针对所有评估基准去重后的数据上进行训练，随后组合成长时程搜索，在单一循环中耦合学习与演化。在MLE-Bench Lite上，以单张RTX 4090（显存上限12 GB）每任务12小时为预算，Frontis-MA1（35B）借助OpenMLE-Evo在基座模型基础上将奖牌平均分（Medal Average）从39.39%提升至60.61%；使用OpenMLE-Evo-Max（基准无关的经验先验和异步搜索）时达到71.21%，超过GPT-5.5 + Codex，逼近GPT-5.6 Sol和2.8T参数的Kimi K3。在留出的NatureBench Lite上，两个组件均具有迁移性：固定框架时，换用训练后的模型将Match-SOTA从50%提升至70%；固定模型时，换用OpenMLE-Evo将其从20%提升至50%。我们发布了模型权重和完整的OpenMLE技术栈，以支持面向RSI的可执行AI4AI的可复现研究。代码：https://github.com/FrontisAI/OpenRSI

English

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-RL), and long-horizon search (OpenMLE-Evo). On this stack we post-train 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. On held-out NatureBench Lite, both components transfer: with the framework fixed, swapping in the trained model raises Match-SOTA from 50% to 70%; with the model fixed, swapping in OpenMLE-Evo raises it from 20% to 50%. We release the model weights and the full OpenMLE stack to enable reproducible research on executable AI4AI toward RSI. Code: https://github.com/FrontisAI/OpenRSI