NVIDIA AVO Reaches 100% on ARC-AGI-3, Demonstrating a Frontier-Level General-Purpose Architecture for Long-Horizon Autonomous Agents
NVIDIA, Friday, August 21st, 2026
NVIDIA's AVO agent system lifted Claude Opus 5 from a 30% baseline to 100% on ARC-AGI-3, showing harness design matters.
NVIDIA reports that its AVO research project reached 100% on ARC-AGI-3. The post argues a frontier language model is only one component of an AI agent, and that the surrounding system, often called a harness, determines how the model receives context and acts over long horizons.
AVO elevated Claude Opus 5 from a 30% model baseline to 100% as part of the complete agent system. NVIDIA presents this as evidence that system design, not model capability alone, unlocks frontier-level long-horizon performance. Authors include Terry Chen and colleagues.