NVIDIA Open-Sources a Pythonic Agent Harness - 2026-07-27
NVIDIA's NOOA release shows how typed state and pass-by-reference can improve agent accuracy while cutting token cost.
In one minute
The details
NVIDIA’s agent harness beats bigger prompts with typed Python state - 10/10
Takeaway: Agent performance is a systems problem: typed state and pass-by-reference beat bigger prompts, but NOOA still executes model-written Python.
What changed: NVIDIA open-sourced NOOA, reporting 82.2% SWE-bench Verified with GPT-5.5 at roughly half the tokens of comparison harnesses.
Sources: developer.nvidia.com: NVIDIA Technical Blog: Six Agent Harness Capabilities, GitHub: NVIDIA Labs Object-Oriented Agents, blogs.nvidia.com: NVIDIA: Open Secure AI Alliance, X/Twitter @NVIDIAAI: NVIDIA AI release thread
More links
blogs.nvidia.com: NVIDIA says Vera speeds selected chip-design workloads by up to 1.5x. (6/10 / context). CPU tuning still controls simulation and verification throughput in the feedback loop behind new AI chips.
Quick feedback
Send what was useful, what to cut or rank lower, what was missed, and how the link list should change.