Robots Atlas>ROBOTS ATLAS

Thinking Machines Lab

ACTIVE
Thinking Machines Lab Inc.
StartupAI Lab
Thinking MachinesTML

American AI startup founded in 2025 by Mira Murati, former CTO of OpenAI; develops foundation models and human–AI interaction tools.

Founded2025United States
HQSan FranciscoUnited States
Total funding$2000000B1 · 2025
ValuationUSD 12 billion🦄Seed round valuation · 2025
thinkingmachines.aiWikipedia
01 · About

Thinking Machines Lab

Founded 2025 · San Francisco, United States

Thinking Machines Lab Inc. is an American artificial intelligence startup founded in February 2025 by Mira Murati, former Chief Technology Officer of OpenAI. The company is headquartered in San Francisco and structured as a public benefit corporation. It employs about 100 people (as of 2026), including former OpenAI, Meta AI and Mistral AI staff. Founding members and key people include John Schulman (Chief Scientist), Lilian Weng and Barret Zoph. In July 2025 the company closed a seed round of about USD 2 billion led by Andreessen Horowitz at a USD 12 billion valuation; investors included NVIDIA, AMD, Cisco, Jane Street and the government of Albania (USD 10 million). The first product was Tinker — an API for fine-tuning language models, announced on 1 October 2025. On 11 May 2026 the company unveiled the native interaction model TML-Interaction-Small (276B MoE, 12B active parameters).

Founders

  • Mira Murati

    Founder and CEO

    Albanian-American engineer, former Chief Technology Officer of OpenAI (2022–2024). Founded Thinking Machines Lab in February 2025.

  • John Schulman

    Chief Scientist

    OpenAI co-founder, author of the PPO (Proximal Policy Optimization) algorithm. Joined Thinking Machines Lab after a brief tenure at Anthropic.

02 · What they do

Domain of activity

3 areas
What they do3Areas
  • Language models
  • Multimodal models
  • AI platforms
03 · Geography

Global presence

HQ San Francisco · 2 regions
USAGlobal
04 · Metrics

Scale & funding

founded 2025 · Seed
2025founded
about 100 (2026)employees
$2Btotal funding
Seedlatest round
earlydevelopment stage
Privatenot listed
activestatus
$2B
total funding
1 · 2025
SeedJuly 2025Andreessen Horowitz (a16z), NVIDIA, AMD, Cisco, Jane Street, Government of AlbaniaSeed round led by Andreessen Horowitz at a USD 12 billion valuation. Albania invested USD 10 million following an amendment to its 2025 state budget.
USD 2 billion
05 · Portfolio

Products & solutions

1 ai models
06 · Relations

Organizational relations

1 strategic relations

Strategic relations

1
  • Compute partner
    NVIDIAMulti-year strategic partnership including an undisclosed NVIDIA investment and deployment of 1 GW of Vera Rubin compute capacity.
    2026 →
07 · Classification

Profile & metadata

4 classification · 3 external links

Classification

  • Statusactive
  • Development stageearly
  • Listed onNoprivate
  • Organization typeStartup+ AI Lab

External links

08 · Related content

Articles

3 articles
Lilian Weng leaves Thinking Machines and returns to OpenAI
News31 July 2026

Lilian Weng leaves Thinking Machines and returns to OpenAI

Lilian Weng, Thinking Machines Lab co-founder and former OpenAI VP of AI safety, left the startup citing her health and returned to OpenAI to lead a team working on recursive self-improvement.

Thinking Machines reveals Inkling — Mira Murati open-weight MoE 975B
News17 July 2026

Thinking Machines reveals Inkling — Mira Murati open-weight MoE 975B

Thinking Machines Lab, Mira Murati startup, released its first AI model — Inkling. An open-weight MoE with 975 billion total parameters, natively multimodal, trained on 45 trillion tokens. The company argues what matters is not the model itself but the ability to customize it.

Thinking Machines debuts: TML-Interaction-Small merges intelligence with real-time responsiveness
News12 May 2026

Thinking Machines debuts: TML-Interaction-Small merges intelligence with real-time responsiveness

Mira Murati's startup has released its first foundational model — TML-Interaction-Small. The 276B-parameter MoE model is trained from scratch to handle audio, video, and text in 200-millisecond micro-turns, without any external dialog-management components.