---
title: "Prime Agent: A Self-Improving RLM Harness That Pushes ARC-AGI-3 to 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","Open-source Framework","Recursive Language Model","Long-horizon Agent"]
---

# Prime Agent: A Self-Improving RLM Harness That Pushes ARC-AGI-3 to 95.5%

## AI commentary

> What makes this paper worth reading is not just another agent framework, but its revelation that the harness itself can become the bottleneck for model capability. For those building agent systems, benchmarks, or autonomous coding tools, Prime Agent's memory-skill-subagent structure is highly instructive — and the 3x improvement on ARC-AGI-3 is worth reproducing firsthand.

## At a glance

Prime Agent is an open-source long-horizon agent harness that uses a persistent IPython REPL, a Continual Harness, and recursive subagents to enable programmatic context processing and test-time compute. It standardizes execution, recovery, verification, and resource accounting, preventing harness failures from becoming model failures. Experiments show 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. In Factorio, refinement enables continuous technology progression and dedicated subagents allow parallelized work. The code is publicly available.

## Key points

- Built on the Recursive Language Model abstraction, Prime Agent uses a persistent IPython REPL for programmatic context processing and test-time compute.
- The Continual Harness preserves histories, memories, skills, prompts, and subagent specifications across trajectories, enabling self-improvement.
- It raises ARC-AGI-3 RHAE Best@1 from 30% to 95.5% and matches or exceeds existing harnesses in coding, GPU-kernel generation, emulation, and nanoGPT speedruns.
- An Agents View lets humans inspect and manage daemon-backed sessions, emphasizing observability and a low-friction interface.

## Full text

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.

---

Verify against the original before citing; curated by Astral Feed: https://feed.astralor.com/items/itm_a0eb6a8f44dd6a7e0b30c4eb