Cadence has unveiled Maestro, a suite of fine-tuned AI models designed to support specialized chip design agents. The system utilizes reinforcement learning and integrates with Nvidia’s Nemotron, allowing customers to perform additional training on their own infrastructure to protect intellectual property.
The company reports that Maestro achieves up to 30% higher quality results and up to 5x greater token efficiency compared to general-purpose foundation models. These gains are attributed to a strategy of balancing EDA tool calls with LLM invocations, optimizing resource usage for specific engineering tasks.
Maestro is scheduled for early access to select customers in Q4 2026. The platform will be compatible with Cadence’s cloud, public clouds, and on-premises environments. Cadence is also collaborating with Nvidia on the OpenShell Reference Architecture to enhance agent security.
This development follows a series of agent-focused initiatives by Cadence, including the acquisition of ChipStack in November 2025 and the deployment of various super agents like ViraStack and InnoStack. Currently, Cadence has about a dozen customer engagements for the Maestro platform.
