---
title: "Prime Agent：自改进的递归语言模型执行框架，将 ARC-AGI-3 成绩提升至 95.5%"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2608.23552v1"
permalink: "https://feed.astralor.com/items/itm_a0eb6a8f44dd6a7e0b30c4eb"
published: "2026-08-24T17:54:19.000Z"
score: 77
channel: "ai"
tags: ["ARC-AGI-3","Prime Agent","Prime Intellect","开源框架","递归语言模型","长时程智能体"]
---

# Prime Agent：自改进的递归语言模型执行框架，将 ARC-AGI-3 成绩提升至 95.5%

## AI 点评

> 这份论文最值得看的地方不是又一个智能体框架，而是它揭示了评测框架本身可能成为模型能力的上限瓶颈。对做智能体系统、评测基准或自主编程工具的人来说，Prime Agent 的“记忆-技能-子代理”结构很有借鉴意义，尤其是 ARC-AGI-3 上三倍多的提升，值得亲自跑一遍确认。

## 速览

Prime Agent 是一个开源的长时程智能体执行框架，通过持久 IPython REPL、Continual Harness 和递归子代理实现程序化上下文处理与测试时计算。它标准化执行、恢复、验证和资源核算，避免框架缺陷污染模型能力测量。实验显示，Prime Agent 将 ARC-AGI-3 RHAE Best@1 从 30% 提升到 95.5%，并在长上下文编码、GPU 内核生成、模拟器构建和 nanoGPT 速通等任务中达到或超过原生与主流框架。在 Factorio 中，细化机制支持持续科技推进，专用子代理实现并行工作。

## 关键点

- Prime Agent 基于递归语言模型（RLM）抽象，用持久 IPython REPL 进行程序化上下文处理和测试时计算。
- Continual Harness 跨轨迹保存历史、记忆、技能、提示和子代理规格，实现自改进与知识沉淀。
- 在 ARC-AGI-3 RHAE Best@1 上从 30% 提升至 95.5%，在多项编程与自主任务中达到或超越现有框架。
- 提供 Agents View 供人类检查和管理守护进程会话，强调可观测性与低摩擦接口。

## 全文

Language models are sequential processors, but long-horizon agency requires external information and computation beyond model weights and active context. Prime Agent is an open-source harness for long-horizon evaluation and coding-agent workflows. A persistent IPython REPL follows the Recursive Language Model abstraction for programmatic context processing and test-time compute, while Continual Harness preserves histories, memories, skills, prompts, and subagent specifications across trajectories. Recursive subagents coordinate through direct agent-to-agent communication, and the Agents View lets humans inspect and manage daemon-backed sessions. Prime Agent standardizes execution, recovery, verification, and resource accounting while leaving strategy construction to the model. This low-friction, expressive membrane prevents harness failures from becoming model failures and pushes measurement toward the model's true maximal underlying capability. Prime Agent raises ARC-AGI-3 RHAE Best@1 from 30% to 95.5% and matches or exceeds native and popular harnesses across long-context coding, GPU-kernel generation, emulator construction, and autonomous nanoGPT speedruns. On Factorio, we find refinement allows for continuous technology progression and dedicated subagents enable parallelized work. Code is available at https://github.com/PrimeIntellect-ai/prime-agent.

---

引用前请核对原文;本文由星际流动整理:https://feed.astralor.com/items/itm_a0eb6a8f44dd6a7e0b30c4eb