Cadence, a major electronic design automation (EDA) company, announced Maestro, a new portfolio of domain-specialized AI models built for its chip design agents, in a press release ahead of earnings. The company claims initial results show up to a 30% improvement in quality of results and up to 5x token efficiency compared with a general-purpose foundation model, which Cadence described as the open-source model Maestro started from before its training. Cadence did not specify what the 30% figure measures; in chip design, quality of results typically refers to power, performance, and area.
The first Maestro models are available to select early-access customers in Q4 2026. Customers can run them on Cadence’s cloud, public clouds, or customer-controlled on-premises environments, and can post-train them on their own proprietary data, designs, and workflows using Cadence’s fine-tuning harness.
Cadence’s approach differs from that of competitor Synopsys, which in September announced GPT-Synopsys, a specialized model that operates on OpenAI-hosted infrastructure without using customer data to train the model. The two EDA giants, announcing their models less than a month apart, now split on where customer design data goes.
Paul Cunningham, senior vice president and general manager at Cadence, told Tom’s Hardware Premium that agent costs have “really started to explode,” adding, “And I think the world realized so.” He said it is a natural step to move more of the agents’ iterating “into the weights of the models themselves.”
Maestro uses Cadence’s AI Super Agents, which “interpret design intent, decompose tasks” and “orchestrate the Maestro portfolio with other leading frontier and open-weight models.” It also employs reinforcement learning, where Cadence’s “deterministic solvers evaluate outputs against engineering objectives, constraints and ground truth,” with results feeding back into the models for improvement.
The models include Nvidia’s Nemotron. Cunningham said customers can run the customer-data training step themselves, potentially “on their own GPUs, on their own network,” and that Nemotron was “giving really solid results and actually improving very, very fast.”
Cadence credits the token savings to balancing its EDA and system design automation “tool calls with LLM invocations.” As an analogy, Cunningham said, “You don’t need to drive to Safeway in the jet engine,” and that the goal is to use “the right model at the right time for the right task.”
For agent security, such as sandboxing, Cadence says it is working with Nvidia on an integration with Nvidia’s OpenShell Reference Architecture, which Nvidia launched last week, naming both Cadence and Synopsys.
Cadence’s agentic AI roadmap includes the ChipStack AI Super Agent, which has been in early access since February 2026, and the ViraStack, InnoStack, and AgentStack agents announced in April 2026. In June 2026, Cadence claimed ChipStack had reached Level 5 autonomy. In September 2026, Rob Knoth, senior group director of strategy and new ventures, teased Maestro’s ingredients: “reinforcement learning loops, domain-specific post-training of open-weight models, model routing.”