---
title: "NVIDIA AVO Reaches 100% on ARC-AGI-3, Demonstrating a Frontier-Level General-Purpose Architecture for Long-Horizon Autonomous Agents"
description: "NVIDIA’s Agentic Variation Operators (AVO) is a general-purpose agent architecture designed to make frontier language models reliable on extended, multistep tasks. It combines tool use and execution with persistent memory, an iterative inspect-plan-implement-evaluate loop, and a supervisor that can redirect work when progress stalls. In a seven-day GPU-kernel optimization study, AVO explored more than 500 directions, produced 40 committed versions, and outperformed cuDNN by up to 3.5% and FlashAttention-4 by up to 10.5% across evaluated configurations. On the ARC-AGI-3 public set, the same architecture achieved a 100.00 RHAE score across 25 environments, completing all 183 levels in 6,624 actions; the post cautions that comparisons with other systems are not controlled ablations."
---

# NVIDIA AVO Reaches 100% on ARC-AGI-3, Demonstrating a Frontier-Level General-Purpose Architecture for Long-Horizon Autonomous Agents

[NVIDIA](https://yomu.fyi/company/nvidia-developer-blog) · Tanya Lenz · Aug 21, 2026

**Type:** Explainer

## Summary

NVIDIA’s Agentic Variation Operators (AVO) is a general-purpose agent architecture designed to make frontier language models reliable on extended, multistep tasks. It combines tool use and execution with persistent memory, an iterative inspect-plan-implement-evaluate loop, and a supervisor that can redirect work when progress stalls. In a seven-day GPU-kernel optimization study, AVO explored more than 500 directions, produced 40 committed versions, and outperformed cuDNN by up to 3.5% and FlashAttention-4 by up to 10.5% across evaluated configurations. On the ARC-AGI-3 public set, the same architecture achieved a 100.00 RHAE score across 25 environments, completing all 183 levels in 6,624 actions; the post cautions that comparisons with other systems are not controlled ablations.

## Context

The work addresses the challenge of building agent systems that can sustain reliable progress on long-running, multistep tasks rather than relying on a single model response. ARC-AGI-3 provides a different test: agents enter unfamiliar environments without instructions, explicit rules, or stated goals and must infer dynamics, objectives, and efficient action sequences.

## Approach / What changed

AVO uses a general-purpose agent loop for inspecting context, planning, making changes, executing tools, evaluating feedback, and preserving state. Persistent memory carries forward implementations, results, outputs, and reasoning, while a supervisor monitors the broader search for stagnation and can redirect the main agent. For ARC-AGI-3, the same agent was connected to a separately reimplemented direct-interaction task interface.

## Takeaways

- AVO’s GPU-kernel optimization run lasted seven days, explored more than 500 optimization directions, and produced 40 committed kernel versions.
- AVO completed all 183 levels in the ARC-AGI-3 public set with a 100.00 RHAE score using 6,624 environment actions.
- The ARC-AGI-3 comparison with VISTA is not a controlled ablation because the systems differ in backend, observations, memory, context management, and other implementation details.

**Tags:** [Architecture](https://yomu.fyi/topic/architecture), [Machine Learning](https://yomu.fyi/topic/machine-learning), [Performance](https://yomu.fyi/topic/performance)

- Source: [NVIDIA](https://developer.nvidia.com/blog/nvidia-avo-reaches-100-on-arc-agi-3-demonstrating-a-frontier-level-general-purpose-architecture-for-long-horizon-autonomous-agents)
- Source URL: https://developer.nvidia.com/blog/nvidia-avo-reaches-100-on-arc-agi-3-demonstrating-a-frontier-level-general-purpose-architecture-for-long-horizon-autonomous-agents
- Ingested by Yomu: 2026-08-29T01:38:07.212Z

[Read original post](https://developer.nvidia.com/blog/nvidia-avo-reaches-100-on-arc-agi-3-demonstrating-a-frontier-level-general-purpose-architecture-for-long-horizon-autonomous-agents)