Mira Murati's Thinking Machines to Launch Open-Source Product Within Months

Mira Murati, founder of AI startup Thinking Machines and former chief technology officer at OpenAI, announced today a new $2 billion round of venture funding. She confirmed that her company's first product will be released in the coming months and will feature a significant open‑source component designed to help researchers and startups building custom AI models.

The news has energized followers who have been anticipating Murati's next move since she left OpenAI in September 2024 as part of a wave of high‑profile researcher and leadership departures. Her announcement comes at an opportune moment, given OpenAI's recent update that its own forthcoming open‑source frontier AI model—still unnamed—has been delayed.

As Murati posted on X:

Thinking Machines Lab exists to empower humanity by advancing collaborative general intelligence.

We are building multimodal AI that aligns with how people naturally interact with the world—through conversation, sight, and the often‑messy ways we collaborate. We are excited to share our first product in the next few months, which will include a major open‑source element useful for researchers and startups creating custom models. Soon, we will also release our best scientific insights to help the research community better understand frontier AI systems.

To speed our progress, we are pleased to confirm a $2B funding round led by a16z, with participation from NVIDIA, Accel, ServiceNow, CISCO, AMD, Jane Street, and others who support our mission.

We are always seeking exceptional talent who learn by doing and turn research into practical tools. We believe AI should amplify individual agency and, in the spirit of freedom, be distributed as widely and equitably as possible. We hope this vision resonates with those who share our dedication to moving the field forward. If so, join us. https://thinkingmachines.paperform.co

Widespread enthusiasm from the Thinking Machines team

Other employees at Thinking Machines have echoed the excitement about the company's product and infrastructure developments.

Alexander Kirillov described the initiative on X as "the most ambitious multimodal AI program worldwide," pointing to rapid progress over the past half‑year.

Horace He, another engineer at the company, emphasized their early efforts to create scalable, efficient tooling for AI researchers. "We're building some of the best research infrastructure out there," he posted. "Research infrastructure is about optimizing both researcher and GPU efficiency, and it's been a pleasure tackling this challenge with the talented team here."

Investor Sarah Wang of a16z likewise shared her enthusiasm for the team's track record and potential. "Thrilled to back Mira Murati and the world‑class team behind nearly every major recent AI research and product breakthrough," she wrote. "From RL (PPO, TRPO, GAE) to reasoning, multimodal systems, Character, and of course ChatGPT! No one is better positioned to advance the frontier."

More about Thinking Machines

According to Murati, the company’s goal is to deliver systems that are not only technically advanced but also adaptable, safe, and widely accessible. Their strategy stresses open science, including public releases of model specifications, technical papers, and best practices, alongside safety protocols such as red‑teaming and post‑deployment monitoring.

As VentureBeat previously reported, Thinking Machines emerged after Murati’s exit from OpenAI in late 2024. The startup now joins a growing number of new players aiming to redefine how sophisticated AI tools are created and shared.

A timely announcement after OpenAI delays its own open‑source foundation model

Murati’s announcement arrives amid heightened interest in open‑access AI, following OpenAI’s move to postpone the launch of its long‑awaited open‑weight model.

The release, originally set for this week, was recently delayed by CEO and co‑founder Sam Altman, who cited the need for more safety testing and further review of high‑risk areas.

As Altman posted on X last week:

We planned to launch our open‑weight model next week.

We are delaying it; we need time to run additional safety tests and review high‑risk areas. We are not yet sure how long it will take.

While we trust the community will create great things with this model, once the weights are out, they can’t be taken back. This is new territory for us and we want to get it right. Sorry for the disappointing news; we are working extremely hard!

Altman acknowledged that releasing model weights is irreversible and stressed the importance of ensuring everything is correct, without offering a new timeline.

First announced publicly by Altman in March, the model was described as OpenAI’s most open release since GPT‑2 in 2019—long before the November 2022 debut of ChatGPT, which ran on GPT‑3.

Since then, OpenAI has concentrated on launching increasingly powerful foundation large language models (LLMs), but kept them proprietary, accessible only via its ChatGPT interface (with limited free interactions) and paid subscribers to its apps such as Sora, Codex, and its platform API. This approach has frustrated many of its initial open‑source advocates as well as former funder and co‑founder turned AI rival Elon Musk, who now heads xAI.

However, the launch of the highly capable open‑source DeepSeek R1 by Chinese firm DeepSeek (an affiliate of High‑Flyer Capital Management) in January 2025 dramatically reshaped the AI model landscape. It quickly soared in usage rankings and app downloads, offering advanced reasoning features once exclusive to proprietary models—for free—along with full customizability, fine‑tuning options, and the ability to run locally for privacy‑conscious users.

Other leading AI providers, including Google, subsequently released similarly powerful open‑source models in an effort to attract users into their ecosystems. It appears OpenAI also felt competitive pressure and began developing its own open‑source alternative.

Altman has characterized the upcoming OpenAI open‑source release as a model with reasoning abilities, emphasizing its role as a foundation for developers to build and fine‑tune their own systems. OpenAI has also held developer feedback sessions in San Francisco, Europe, and Asia‑Pacific, indicating the model was still being refined.

Other OpenAI staff, such as Aidan Clark, reiterated on X that the model is highly capable but must meet stringent safety standards.

The postponement, combined with limited technical details and a clear schedule, suggests the project remains in a cautious, internally focused stage.

This delay has created an opening in the developer ecosystem—one that Thinking Machines now appears ready to fill with a clearer timeline and a public commitment to openness.

With OpenAI’s open‑weight model now on hold, Thinking Machines’ announcement of a definite schedule and an open‑source component could redirect developer interest. By signaling public readiness and a dedication to openness, the company is not only asserting a role in the competitive AI frontier but also responding to developer demands for transparent, adaptable tools.