Mira Murati's Thinking Machines Unveils Real-Time 'Interaction Model' — 0.40-Second Response Speed, No Public Access Yet

What Actually Happened

On May 11, 2026, Thinking Machines Lab announced interaction models — AI that processes audio, video, and text input while simultaneously generating a response. The technology operates in full duplex, similar to a phone call rather than back-and-forth messaging.

The company's flagship demo model, TML-Interaction-Small, achieves a response time of 0.40 seconds, according to TechCrunch. Thinking Machines claims this outperforms comparable models from OpenAI and Google.

What the Model Actually Does

Today's AI models process information sequentially — they finish generating output before accepting new input. Thinking Machines says interaction models eliminate that constraint by continuously ingesting audio, video, and text while responding in real time across all three.

Demos include real-time speech translation, detecting animal mentions in stories, and identifying poor posture. Whether these capabilities work outside controlled demonstrations remains unclear.

What You Can't Do Yet

The model is not available to the public. Thinking Machines plans a limited research preview in coming months, with wider release later in 2026. This announcement precedes any actual product availability.

Lab benchmarks often diverge from real-world performance with millions of users. TechCrunch noted: "Whether the real-world experience lives up to the technical claims is something we won't know until people can actually use it."

The Company Behind It

Thinking Machines Lab was founded in February 2025 by Mira Murati and OpenAI alumni including John Schulman (co-founder and chief scientist), Barrett Zoph (former VP of Research), Lilian Weng (former VP), Andrew Tulloch, and Luke Metz.

In July 2025, Andreessen Horowitz led a funding round raising $2 billion at a $12 billion valuation. Other investors included Nvidia, AMD, Cisco, Accel, and Jane Street. This was the largest seed round in history.

The company operates as a public benefit corporation with Murati holding controlling board votes. Founding shareholders receive votes weighted 100 times greater than regular shareholders.

Notably, the government of Albania — Murati's country of origin — invested $10 million in the round, requiring an amendment to Albania's 2025 national budget.

The Talent Exodus

Thinking Machines has experienced significant departures among senior leadership.

Co-founders Barrett Zoph and Luke Metz left in January 2026 and returned to OpenAI. Andrew Tulloch joined Meta Superintelligence Labs. Multiple other key hires have defected to Meta, as The Verge reported.

When founders depart to rejoin their previous employer, it signals structural problems worth examining.

Questions About the Business Model

Tech press coverage came primarily from The Verge, TechCrunch, and WIRED, with limited scrutiny of the underlying business case.

Several concerns warrant attention:

First, $2 billion in funding for a company with zero revenue-generating products at the time of investment is extraordinary. Tinker, the first product launched in October 2025, is a developer fine-tuning API — not a consumer offering. The path to profitability remains unclear.

Second, the weighted voting structure means outside investors including Nvidia and Andreessen Horowitz hold minimal governance influence despite billion-dollar investments.

Third, the AI industry has a history of well-funded startups that demonstrated compelling prototypes but failed to deliver commercial products. Skepticism is reasonable.

Fourth, Albania's government investment and budget amendment for a $10 million stake in a private American company is unusual and deserves scrutiny as a geopolitical and governance matter.

The Bottom Line

Murati's team has credible technical expertise. The interaction model concept is genuinely interesting if full-duplex AI functions as described.

However, this remains a demo from a company losing co-founders, holding $2 billion in venture funding, with no public product and no proven revenue. Widespread availability and real-world performance will determine whether this announcement merits the attention it has received.