Mira Murati, founder of Thinking Machines, says that her company will release a new product with a’significant component of open source’ in a few months

Subscribe to our weekly newsletters and get the latest news on enterprise AI, data security, and data analytics. Subscribe Now Mira Murati is the founder of AI startup Thinking Machines, and former Chief Technology Officer of OpenAI. It is exciting news for those who have been waiting for Murati’s new venture, as she left OpenAI in September of 2024, part of a wave that saw high-profile researchers and leaders leave. This news also comes at a good time, given that OpenAI recently announced that its own open source frontier AI model, which has yet to be named, would be delayed.

Murati Wrote: X

: “Thinking Machines Lab exists in order to empower humanity by advancing collaborative general Intelligence.

Our multimodal AI is designed to work with the way you naturally interact with the outside world – by conversation, by sight, and through the messy ways we collaborate. In the next few months, we will be able share our first product. It will include an important open source component that can be useful to researchers and startups who are developing custom models. We’ll soon share our best science with the research community to help them better understand frontier AI systems.

We’re pleased to confirm that, in order to accelerate our progress, a16z, NVIDIA Accel, ServiceNow CISCO AMD Jane Street, Jane Street, and others who share our mission have raised $2B.

Our team is always on the lookout for exceptional talent who can learn by doing and turn research into useful products. We believe AI should be a natural extension of the individual’s agency, and distributed as widely as possible in the spirit freedom. We hope that this vision resonates among those who share our commitment in advancing the field. If so, join us. https://thinkingmachines.paperform.co“

The team at Thinking Machines is ecstatic

Many other Thinking Machines employees are also excited about the progress of the product and infrastructure.

Alexander Kirillov (19459061) described it on X, noting rapid progress in the past six months.

Horace He (19459061), another engineer at the firm, highlighted their early works on scalable, efficient tools for AI researchers. He posted: “We’re building the best research infrastructure around.” “Research infrastructure is about optimizing GPU efficiency for both researchers and investors. It’s been a pleasure to work with the other great folks here.”

Sarah Wang, of a16z (19459061), also expressed her excitement about the team and its potential. She wrote, “Thrilled that I can support Mira Murati and her world-class team who are behind every major AI research and product breakthrough in recent years.” “RL (PPO TRPO GAE), reasoning multimodal, Character and of course ChatGPT!” No one is better placed to advance the frontier.”

More on Thinking Machines (19659024) According to Murati the company is aiming to deliver systems which are not only technically capable, but also adaptable and safe. Their approach emphasizes open-science, including the public release of model specs, technical documents, and best practices along with safety measures like red-teaming, and post-deployment monitor.

According to VentureBeat, Thinking Machines was formed after Murati left OpenAI at the end of 2024. The company is one of many new entrants that are aiming to change the way advanced AI tools and services are developed and distributed.

An announcement that is well-timed following OpenAI delaying its own open-source foundation model

This announcement comes amid increased interest in open-access AI following OpenAI’s decision to postpone the release of its much-anticipated open-weight model.

Originally scheduled for this week’s release, the planned release was delayed. Sam Altman, CEO and cofounder of the companyhas recently postponed the launch date. He cited a need for more safety testing and a review of high-risk zones.

According to Altman Last week, I wrote on X:

We planned to launch our open weight model next week.

We are delaying the launch; we need to run additional safety testing and review high-risk zones. We are not sure how long this will take.

While we are confident that the community will create great things using this model, weights cannot be pulled back once they have been released. This is a new model for us, and we want it to be perfect. We are working hard, sorry for the bad news.Altman acknowledged that model weights are irreversible and stressed the importance of getting them right without providing a new timetable.

Altman first announced the model publicly in March. The model was billed as OpenAI’s most open release in 2019 since GPT-2 — long before the Release of ChatGPT powered with GPT-3 in November 2022

OpenAI has been focusing on releasing more powerful foundation large-language models (LLMs) but has kept them proprietary. They are only accessible via its ChatGPT interface, with limited interactions for users of the free tier, and to paid subscribers. This angered many of its original open source supporters, including former funder, co-founder, and AI rival Elon (who is currently leading xAI). The launch of the powerful DeepSeek R1 open source model by Chinese firm DeepSeek in January 2025 (an offshoot from High-Flyer Capital Management), completely upended the AI market. It offered advanced AI reasoning previously reserved for proprietary models, for free. And it ran locally without web servers, for those who are concerned about privacy.

Google and other major AI providers were then motivated to release similar powerful open source AI model in an attempt to attract users to their ecosystems. It appears that OpenAI “felt” the competitive heat and was also moved to develop its own open-source rival. Altman described the new OpenAI release as a reasoning model and stressed its use as a basis for developers to build their own systems. OpenAI also held feedback sessions in San Francisco, Europe and Asia-Pacific for developers to gather input. This signaled that the model is still being refined.

OpenAI employees include Aidan Clarkreiterated on X, that the model has a strong capability but must also meet a high bar in terms of safety. This pause, coupled with the lack technical detail and clear dates suggests that the initiative is still in a cautious phase.

This delay has created an opening in the developer’s ecosystem, one that Thinking Machines is now poised to fill with a clearer timeline and a public commitment towards openness.

OpenAI’s open weight model is now in limbo. Thinking Machines’ announcement of a clear timeline, and the inclusion of an open source component, could reshape developers’ attention. By committing to openness and signaling its public readiness, the company is not just staking a position on the competitive frontier of AI but also meeting developer demand for customizable, transparent tools.

Daily insights into business use cases from VB Daily

Want to impress your boss? VB Daily can help. We provide you with the inside scoop on what companies do with generative AI. From regulatory shifts to practical implementations, we give you the insights you need to maximize ROI.

Read our privacy policy

Thank you for subscribing. Click here to view more VB Newsletters.

An error occured.