Skip to main content

Thinking Machines Lab

Category: Foundation Models / LLMs

San Francisco PBC AI lab (ex-OpenAI CTO Mira Murati) building open-weights multimodal foundation models (Inkling) and the Tinker platform so organizations can customize frontier AI rather than only rent closed models. Thinking Machines Lab was founded in 2025. The company is led by Mira Murati. Based in San Francisco, USA. Total funding raised: $2.0B. Latest round: Seed ($2.0B, Jul 2025). Key investors include Andreessen Horowitz (a16z), Nvidia, AMD, Cisco, Jane Street, Accel, ServiceNow, Lightspeed Venture Partners.

AMW Analysis

Thinking Machines Lab, an AI research lab founded in 2024, is building collaborative general intelligence and tools to democratize frontier AI customization. The company raised $2.0B in seed funding in July 2025 from investors including Andreessen Horowitz, Nvidia, AMD, and Cisco. In July 2026, the startup released its first major product, Inkling, an open-weight Mixture-of-Experts model with 975 billion total parameters (41 billion active) and a 1-million-token context window, trained on 45 trillion tokens across text, image, audio, and video data. The model weights were published publicly, positioning Inkling as a Western alternative to leading open-source models. Prior to this launch, in May 2026, Thinking Machines Lab previewed interaction models that process continuous human communication through camera and microphone inputs, and published a technical vision for full-duplex real-time conversational AI. The company’s trajectory shows a shift from early technical previews to a concrete open-weight product release, emphasizing model customization and human-in-the-loop interaction.

AMW analysis, generated from 8 tracked news signals.

Founded
2025
Headquarters
San Francisco, USA
Total funding
$2.0B

Value proposition

Democratizes access to frontier AI by providing high-level abstractions that automate the complexity of distributed training and fine-tuning, removing the GPU infrastructure bottleneck.

Products and solutions

Inkling: First in-house open-weights multimodal MoE foundation model (975B total / 41B active params, up to 1M context; Apache 2.0), released Jul 15 2026; Inkling-Small preview (12B active)., Tinker: Cloud training/fine-tuning API for customizing open-weight models (incl. Inkling) with LoRA; launched Oct 2025., Interaction Models: Research preview of multimodal collaborative AI (voice/vision, real-time human-AI interaction)., Open Science: Technical blogs, papers, code, Hugging Face weights, and research grants.

Unique value

Focuses on 'Collaborative General Intelligence' (CGI)—AI designed specifically for human-machine synergy rather than replacement—operating under a Public Benefit Corporation structure led by former OpenAI CTO Mira Murati.

Target customer

AI researchers, developers, startups, and enterprise organizations requiring custom frontier model capabilities without managing complex infrastructure.

Industries served

Artificial Intelligence, Enterprise Software, Academic Research & Higher Education, Cloud Computing & Infrastructure

Technology advantage

Combines a 'star-studded' team of ex-OpenAI researchers with massive capital ($2B seed) to provide proprietary distributed training abstractions that simplify frontier model development for non-specialist organizations.

How they differentiate

PBC focused on human-AI collaboration and customizable open-weights models (vs closed frontier APIs). Stack: Tinker fine-tuning platform + own open-weights multimodal models (Inkling family) so orgs can own/adapt weights rather than rent one-size-fits-all models from OpenAI/Anthropic/Google.

Main competitors

OpenAI, Anthropic, Safe Superintelligence (SSI)

Key partnerships

Andreessen Horowitz (a16z) - Lead Seed Investor, Nvidia - multiyear strategic partnership (Mar 2026): significant additional investment + ≥1GW Vera Rubin compute; seed co-investor, Strategic/seed partners: AMD, Cisco, ServiceNow, Jane Street, Accel, Lightspeed, Inkling ecosystem: Together AI, Fireworks, Modal, Databricks, Baseten, Hugging Face, SGLang/vLLM partners, Academic/early users: Stanford, Princeton

Notable customers

Bridgewater Associates (Tinker fine-tuning of open models for financial reasoning), Stanford University, Princeton University

Major milestones

Founded/launched February 2025 by Mira Murati with OpenAI alumni (Schulman, Zoph, Weng, Tulloch, Metz), $2B Seed led by a16z at $12B valuation (Jul 2025), Launched Tinker fine-tuning API (Oct 2025), Co-founders Zoph & Metz (and Schoenholz) returned to OpenAI; Soumith Chintala named CTO (Jan 2026), Nvidia multiyear partnership: significant investment + ≥1GW Vera Rubin capacity (Mar 2026), Released first open-weights model Inkling (975B MoE) and Inkling-Small preview (Jul 15 2026)

Growth metrics

Record $2B seed at $12B valuation (Jul 2025); team scaled to ~200 (mid-2026) after Jan 2026 talent churn; shipped Tinker (Oct 2025) and first open-weights foundation model Inkling (Jul 2026); Nvidia multiyear compute + capital partnership (Mar 2026). Follow-on raise talks (~$50B valuation) reported late 2025 but not confirmed closed.

Market positioning

Frontier AI research lab and high-end enterprise infrastructure provider

Geographic focus

North America (San Francisco), Global

Patents and IP

No registered patents disclosed as of latest update; the company emphasizes a culture of 'open science' alongside proprietary software abstractions.

About Mira Murati

Former Chief Technology Officer at OpenAI (2018–2024) where she led the development of ChatGPT, GPT-4, DALL-E, and Sora. Previously served as Senior Product Manager at Tesla (2013–2016) leading Model X development, and VP of Product and Engineering at Leap Motion.

Latest news about Thinking Machines Lab

More Foundation Models / LLMs companies

Official website: