Thinking Machines Lab, Explained

The company behind Inkling raised the largest seed round in tech history and bets against one-size-fits-all AI. Who they are, what they ship, and where it is heading.

Thinking Machines: The Anti-Frontier Frontier Lab

Thinking Machines Lab occupies a strange position in the AI race: founded by the person who shipped ChatGPT, funded like a frontier lab, yet explicitly not competing for the strongest-model crown. Thinking Machines Lab's stated mission is building AI that "extends human will and judgment" — in practice, a bet that customized models beat general ones for real products, and that the money is in the tooling.

The Thinking Machines Lab strategy has three visible pillars. Inkling AI — open-weights models (Apache 2.0) designed as customization bases rather than benchmark champions. Tinker — the managed fine-tuning platform where that customization happens, and the likeliest revenue engine. Connectionism — research previews exploring interactive human-AI collaboration. Thinking Machines releasing Inkling's full weights while charging for Tinker mirrors the classic open-core playbook, applied to foundation models.

Thinking Machines Lab: Company Timeline

Feb 2025
Mira Murati founds Thinking Machines Lab after leaving OpenAI, where she was CTO
Jul 2025
Raises a record ~$2B seed round; investors include NVIDIA, a16z, and AMD
Oct 2025
Launches Tinker, a managed fine-tuning platform for open-weights models
Early 2026
Publishes Connectionism research previews on interactive AI collaboration
Apr 2026
Previews internal training milestones; team grows past 100 researchers
Jul 15, 2026
Releases Inkling, its first from-scratch model — 975B params, open weights, Apache 2.0

Thinking Machines Lab timeline compiled from public reporting and official announcements; early dates approximate.

Why Thinking Machines' Open-Weights Bet Might Work

The economics of frontier AI favor incumbents, and Thinking Machines knows it: training runs cost billions and the strongest closed model captures most consumer demand. Thinking Machines sidesteps that fight. By making Inkling AI free to download and monetizing Tinker, it aligns with the fastest-growing segment — companies that want their own model for a vertical product, not a general assistant.

The risks are equally clear: DeepSeek, Qwen, and GLM give away strong open weights too, and third-party fine-tuning services already exist. Thinking Machines' differentiators are the multimodal + 1M-context combination, the day-one managed pipeline, and — not to be discounted — the credibility of a team that built ChatGPT. Whether that is worth a ~$2B seed valuation depends entirely on Tinker adoption over the next year — the metric to watch for Thinking Machines Lab in 2027.

Thinking Machines Lab FAQ

What is Thinking Machines Lab?
An AI research company founded in February 2025 by Mira Murati, former CTO of OpenAI. Its stated mission is building AI that extends human will and judgment, with a focus on customization — letting anyone shape models rather than consuming one-size-fits-all AI.
Who is Mira Murati?
Mira Murati led development of ChatGPT, DALL-E, and GPT-4 as OpenAI's CTO, and briefly served as interim CEO in November 2023. She left OpenAI in late 2024 and founded Thinking Machines Lab a few months later.
Who has invested in Thinking Machines Lab?
The lab raised a record seed round of roughly $2 billion in mid-2025, with backers including NVIDIA, Andreessen Horowitz, and AMD — one of the largest early-stage rounds in tech history.
What products does Thinking Machines Lab have?
Three public pillars: Inkling (the open-weights model family), Tinker (a managed fine-tuning platform), and Connectionism (research previews on interactive AI collaboration), plus published research.
How is Thinking Machines different from OpenAI or Anthropic?
Strategy: instead of racing for the strongest closed frontier model, it releases open weights designed for customization and monetizes the tooling (Tinker) around them — betting that adapted models beat general ones for real products.