Nvidia says its general-purpose coding agent system AVO scored 100% across all 25 environments in the ARC-AGI-3 public set, completing all 183 levels (Terry Chen/NVIDIA Technical Blog)
AVO’s result turned a 30% Claude Opus 5 baseline into a perfect public ARC-AGI-3 run through agent-system design.
Nvidia says AVO completed all 183 levels across 25 public environments without instructions, explicit rules, or stated goals. The system continuously inspects, plans, implements, evaluates, and uses memory plus execution feedback to carry progress across long tasks. Nvidia frames the result as evidence that the agent harness mattered as much as, or more than, the underlying model. Techmeme's note
Nvidia says AVO completed all 183 levels across 25 public environments without instructions, explicit rules, or stated goals. The system continuously inspects, plans, implements, evaluates, and uses memory plus execution feedback to carry progress across long tasks. Nvidia frames the result as evidence that the agent harness mattered as much as, or more than, the underlying model. Techmeme's note
score 7