Nvidia just showed that the harness, not the AI model, is now the real hero
techcrunch ·
Nvidia published research suggesting that the harness, or scaffolding around an AI model, matters more than the model itself for long-horizon tasks. By using a custom harness with strong memory handling and a supervisor component, researchers achieved a 100% score on the ARC-AGI-3 benchmark with Claude Opus 5, whereas the same model scored only 30% without the harness. Long-horizon tasks require many decisions over time, and models often fail or misbehave when left to string decisions together on their own. OpenAI also found that small harness tweaks tripled its models' scores, but none reached 100%. Nvidia's harness, called Agentic Variation Operators (AVO), includes a supervising agent that nudges the main agent when it gets stuck. The company argues that open harnesses give users more control and are key to improving accuracy and security in agentic systems.
techcrunch