New
AI & RoboticsMore in AI & Robotics→
Nvidia's SoL-Pi Cuts Coding Agent Token Usage Nearly in Half by Optimizing the Harness
Key Takeaways
- Nvidia's SoL-Pi optimizes the harness, the control layer between a coding agent and its environment, rather than the model itself.
- Token usage falls by up to 49 percent with roughly unchanged performance, according to the researchers.
- The system explored 152 directions across 535 executable environments, generating over 3,000 runs and 60,000-plus agent-environment interactions.
- Task mix included 495 GitHub issue-pull-request pairs and 40 synthetic test cases.
DE
DT Editorial Team··via the-decoder.com