A star AI company valued at $12 billion fired its first CTO just a few years after its founding.
Just now, Mira Murati, former CTO of OpenAI and founder of Thinking Machines Lab, posted a strongly worded statement on social media:
We have parted ways with Barret Zoph.
Soumith Chintala will assume the role of Chief Technology Officer (CTO) at Thinking Machines. He is a talented and experienced leader who has made significant contributions to the field of artificial intelligence over the past decade and is a core member of our team. We are incredibly excited and excited about his new role.
This wave of personnel changes is more intense than expected.
According to The Verge reporter Kylie Robison, two sources revealed that Barret Zoph was dismissed due to unethical conduct. Mira announced the news at today's all-hands meeting, showing no mercy to her longtime colleague.
According to the latest report from Wired magazine, a source close to Thinking Machines revealed that Zoph had leaked confidential company information to competitors.
The timeline of events is quite intriguing.
Wired magazine obtained an internal memo from Fidji Simo, CEO of OpenAI's Applications division, to employees. Zoph had told Mira on Monday that he was considering leaving the company. He was officially fired on Thursday. This speed makes it seem as if Mira preemptively struck, not giving him a chance to resign voluntarily.
Even more dramatically, Zoph, along with co-founder Luke Metz and employee Sam Schoenholz, are reportedly rejoining OpenAI.
Yes, you read that right. He left OpenAI to start his own business, but after many twists and turns, he's coming back.
Furthermore, Simo specifically mentioned in the memo that "OpenAI's concerns about Zoph are inconsistent with Murati's." Regardless, this is a significant positive for OpenAI, as their VP of Research, Jerry Tworek, recently resigned, and Zoph's return is a perfect replacement.
In fact, Barret Zoph is no nobody.
He rose through the ranks from Google Brain to become the VP of Post-Training at OpenAI, making core contributions to star products like ChatGPT, GPT-4, and GPT-4o. At the May 2024 launch event, he demonstrated GPT-4o's real-time visual problem-solving capabilities, which caused quite a stir at the time.
Barret Zoph
He spent 6 years at Google Brain, focusing on large model sparsity, AutoML, and neural architecture search.
After joining OpenAI in 2022, he co-led the post-training team with John Schulman, responsible for key tasks such as model alignment, tool usage, and multimodal computing. He later left OpenAI with Mira, Bob McGrew, and other senior executives to start his own business, and was considered at the time to be one of the most promising startups within the "OpenAI ecosystem." Unexpectedly, he was ousted so quickly due to misconduct.
Soumith Chintala, who succeeded him, is known as the father of PyTorch. He announced his departure from Meta two months ago, saying that he wanted to "do something small and new that makes him uncomfortable" and did not want to be tied to a single project for life like Linus.
Soumith Chintala
After joining Facebook AI Research in 2013, he rose from L4 software engineer to vice president. In 11 years, he built PyTorch from scratch to an adoption rate of over 90% in the AI field, supporting training with exascale computing power.
Two months after officially announcing his departure from Meta, just one day later, he announced that he was joining Thinking Machines Lab, emphasizing a return to pure technological exploration. He will work alongside former OpenAI core members such as John Schulman, Alec Radford, and Bob McGrew to explore next-generation general intelligent systems.
Speaking of which, Thinking Machines Lab, which is currently in the eye of the storm, has quite a background. Founder Mira Murati previously served as CTO at OpenAI, overseeing key projects such as ChatGPT, GPT-4, and DALL-E. She also briefly served as interim CEO during the boardroom turmoil in November 2023.
Mira Murati
Thinking Machines Lab's mission is to create "understandable, customizable, and more general" multimodal AI, enabling humans to interact with models through natural dialogue, vision, and collaboration. Research focuses on rapid adaptation to single or few samples, and the lab is committed to regularly publishing code, technical blogs, and datasets to promote an open science culture.
The founding team has more than 30 people, about two-thirds of whom come from OpenAI, making it a very impressive lineup.
In addition to Mira and the current CTO Soumith, there are former OpenAI Chief Scientist John Schulman, DALL-E co-founder Alec Radford, Infrastructure VP Bob McGrew, Alignment Research Core member Lilian Weng, and others.
The funding aspect is even more extravagant. In July 2025, it completed a $2 billion seed round, led by a16z, with Nvidia, AMD, Accel, Cisco, Jane Street and others participating, valuing the company at $12 billion, setting a record for the largest seed round in Silicon Valley history.
Of course, publicly available information alone cannot fully reconstruct what actually happened inside. Based on known clues, OpenAI and Thinking Machines Lab have conflicting accounts, and upon closer examination, the power struggles, conflicts of interest, and differences in stances behind this are likely far more complex than they appear on the surface.
This article is from the WeChat official account "APPSO" , author: APPSO, and published with authorization from 36Kr.
