NVIDIA Coding Agent AVO Scores 100 Percent on ARC-AGI Benchmark

21 Aug 2026, 9:37 pm · 22d ago · 1 min read · OfficeChai

NVIDIA announced that its specialized coding agent, named AVO, achieved a 100 percent score on the ARC-AGI benchmark, marking a major milestone in artificial intelligence reasoning. The ARC-AGI benchmark is widely recognized as a rigorous test for evaluating general intelligence and novel problem-solving capabilities beyond traditional pattern recognition. While mainstream large language models have consistently struggled with complex reasoning puzzles on the ARC benchmark, NVIDIA's agent-based coding approach successfully solved all test cases. This breakthrough demonstrates the potential of combining programmatic problem decomposition, autonomous tool use, and specialized code generation architectures to conquer complex abstract reasoning challenges.