---
title: "Frontis-MA1：面向机器学习工程递归自我改进的AI4AI模型"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2607.28568v1"
permalink: "https://feed.astralor.com/items/itm_d480d7554e0a1995f9731b1a"
published: "2026-07-30T17:34:01.000Z"
score: 77
channel: "ai"
tags: ["AI4AI","Frontis-MA1","MLE-Bench Lite","OpenMLE","递归自我改进"]
---

# Frontis-MA1：面向机器学习工程递归自我改进的AI4AI模型

## AI 点评

> 这篇论文把“自我改进”从理念变成了可跑的工程系统，对研究者和平台开发者都很有价值。虽然评测任务规模还不大，但开源权重和全栈框架意味着社区可以在此基础上继续推进。如果你关注 AI 自主进化或 Agent 自我优化，这篇值得读。

## 速览

本文提出 OpenMLE，一个面向机器学习工程递归自我改进的开源全栈系统，并训练 35B 参数的 Frontis-MA1 模型。该模型围绕 Draft、Improve、Debug、Crossover 四个原子操作进行后训练和推理，与长程搜索结合实现学习与进化闭环。在 MLE-Bench Lite 上，Frontis-MA1 将 Medal Average 从 39.39% 提升至 60.61%，OpenMLE-Evo-Max 达 71.21%，超过 GPT-5.5+Codex，接近 GPT-5.6 Sol 与 Kimi K3。模型权重与系统已开源。

## 关键点

- 提出 OpenMLE 全栈开源系统，包括可验证执行环境 OpenMLE-Gym、算子学习 OpenMLE-RL 和长程搜索 OpenMLE-Evo。
- Frontis-MA1（35B）通过执行反馈的 SFT 和 RL 训练四个原子程序进化算子，并与长程搜索耦合。
- 在 MLE-Bench Lite 上，Frontis-MA1 将 Medal Average 从 39.39% 提升至 60.61%，OpenMLE-Evo-Max 达 71.21%，超越 GPT-5.5+Codex，接近 GPT-5.6 Sol 和 Kimi K3。
- 在 NatureBench Lite 上验证了模型与框架的迁移性，并开源模型权重和全部代码。

## 全文

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

---

引用前请核对原文;本文由星际流动整理:https://feed.astralor.com/items/itm_d480d7554e0a1995f9731b1a