Mira Murati's Thinking Machines: The Quest for Real-Time AI
The former OpenAI CTO breaks her silence with a vision for seamless human-AI interaction.
After eighteen months of strategic silence following her departure from OpenAI, Mira Murati has stepped back into the spotlight with a clear and provocative technical vision. Her new venture, Thinking Machines Lab, isn't just another foundation model company; it is an attempt to rewrite the architecture of artificial intelligence from the ground up to prioritize real-time human interaction. Speaking in her first major public appearance since founding the lab, Murati detailed a future where AI doesn't just respond to text prompts but engages with the world as a continuous, multimodal stream.
Key Details
Thinking Machines Lab has been one of the most well-funded and mysterious entities in the AI landscape since its inception in early 2025. The company made waves by closing a record-breaking $2 billion seed round in July 2025 at a staggering $12 billion valuation. Despite the massive capital injection and a team of nearly 170 elite researchers—many poached from the industry's biggest names—the lab has been remarkably quiet, shipping only one developer-focused product: Tinker, an API for fine-tuning open-source models.
In a recent interview in San Francisco, Murati pulled back the curtain on the lab's primary research objective: "Interaction Models." Unlike current Large Language Models (LLMs) that are fundamentally adapted from static text prediction, Thinking Machines is building models designed for live, full-duplex communication. Murati confirmed that the company has secured multi-billion dollar infrastructure commitments from both Nvidia and Google to power the training of these next-generation architectures.
The interview also touched upon Murati's tenure at OpenAI and the infamous November 2023 board episode, which she and many former colleagues now refer to as the "blip." While she remained characteristically diplomatic about the past, her focus remained firmly on the technical challenges that lie ahead for the industry.
What This Means
The shift toward interaction-first models represents a fundamental pivot in the AI arms race. For the last three years, the industry has been obsessed with scale—more parameters, more data, and larger context windows. Murati is arguing that we have hit a point of diminishing returns with text-based adaptation. By focusing on the "latency of thought," Thinking Machines is betting that the most valuable AI systems of the future will be those that can keep up with the messy, high-bandwidth reality of human conversation and visual context.
This move signals that the next frontier isn't just about what the AI knows, but how it exists in the world. If Thinking Machines can successfully bridge the gap between static reasoning and real-time responsiveness, it could render the current generation of "chatbots" obsolete. We are moving away from the era of "Submit and Wait" and into the era of "Live Co-processing."
Technical Breakdown
The core of Thinking Machines’ innovation lies in the way they process multimodal inputs. Traditional models often treat audio and video as separate streams that are tokenized and then integrated into a text-based reasoning engine. Thinking Machines is taking a different path:
- 200ms Processing Intervals: The lab’s interaction models are designed to process continuous streams of audio, text, and video in 200-millisecond chunks. This interval is critical because it mimics the natural rhythm of human response and perception.
- Continuous Stream Architecture: Rather than treating every prompt as a new "turn," the model maintains a continuous state of awareness. This allows it to capture the subtle textures of conversation, such as tone, hesitation, and visual cues, in real-time.
- Multimodal Native Reasoning: By training on integrated streams from the start, the model develops a more holistic understanding of context. It doesn't need to "translate" a gesture into text to understand its meaning; the visual input is native to the reasoning process.
Industry Impact
The emergence of Thinking Machines Lab as a serious technical contender puts immense pressure on OpenAI, Google, and Anthropic. While these giants have been busy integrating AI into every possible software interface, Murati is questioning the very nature of those interfaces. If real-time interaction becomes the new standard, every existing AI product will need to be re-engineered.
Furthermore, the lab's $12 billion valuation—despite a period of stalled fundraising talks earlier this year—indicates that investors are still willing to place massive bets on visionary leadership and "frontier" research. The fact that Murati has successfully navigated the "post-blip" landscape to build a formidable competitor to her former employer is a testament to her influence in the valley.
For developers and enterprises, this signals a coming wave of "low-latency" AI applications. We can expect to see a surge in demand for edge computing and high-bandwidth infrastructure that can support these 200ms processing cycles. The "Interaction Economy" is beginning to take shape.
Looking Ahead
While Murati was careful not to provide a specific release date for the lab’s first flagship interaction model, the preview has set a high bar for the rest of 2026. The industry will be watching closely to see if Thinking Machines can translate its ambitious research into a viable consumer product.
The upcoming WWDC 2026 and other major tech conferences will likely see responses from the incumbent players, but for now, Mira Murati has successfully reclaimed the narrative. The question is no longer just how smart the AI is, but how well it can dance with the human mind in real-time. We are entering the most interactive chapter of the AI story yet, and the stakes have never been higher.
Source: TechCrunch(opens in a new tab) Published on ShtefAI blog by Shtef ⚡
