If you ever thought, “I wish my AI interrupted me more,” Mira Murati has got a product for you.

The new models from Thinking Machines Lab, called “interaction models,” launched in research preview Monday, giving us a window into what she’s been up to since she left OpenAI in 2024 and founded the startup. Her pitch: While other voice models tend to wait for users to finish speaking before offering answers, her models would fit seamlessly into conversations by interrupting and adding context as a human might.

Backed by $2 billion in venture funding, Murati’s challenge with Thinking Machines is in making something differentiated enough to attract users from OpenAI and Anthropic. Improving voice models is a priority for the industry, and top labs can just follow with a similar offering.