Who is Mira Murati?
Mira Murati is cofounder and chief executive of Thinking Machines Lab. She previously served as OpenAI’s chief technology officer. [1]
Born in Albania, she studied at Colby College and Dartmouth through a dual-degree programme, earning a BA and a Bachelor of Engineering. [5]
International education and engineering experience
Pearson College UWC in Canada identifies her as a member of its 2007 class and a UWC Davis scholarship recipient. This period preceded her US university education and placed her within an international educational community. [6]
Dartmouth identifies her with its engineering class of 2012. As a student she participated in Dartmouth Formula Racing, building racing cars in Allyn Laboratory. [14]
Tesla, Leap Motion and the move into AI
At Tesla she worked on designing, developing and launching the Model X. She subsequently led product and engineering teams at Leap Motion, which developed virtual- and augmented-reality interfaces. She also gained experience in aerospace programmes. [5]
OpenAI gave her new responsibilities as senior vice president of research, product and partnerships in December 2020, following her work on the API rollout. [7]
Her appointment as CTO was announced on 5 May 2022. The company highlighted her coordination of research, product and partnerships for DALL·E’s release. The role combined technical and organisational responsibilities necessary to turn research into usable systems. [8]
Responsibilities at OpenAI
As CTO she oversaw teams working on ChatGPT and DALL·E. In 2023 she publicly explained that conversational models could invent facts, and that dialogue with users, feedback and regulators’ involvement should form part of their development. [2]
Dartmouth credits her leadership with products including GPT-4, Codex, Sora and the o-series reasoning models.[1]
She is among the many contributors credited in the 2024 GPT-4o System Card. The document describes safety evaluations and limitations of a multimodal system. This kind of publication reflects collective research, evaluation and implementation work alongside technical leadership decisions. [23]
Interim leadership and departure from OpenAI
On 17 November 2023, the board appointed her interim CEO following Sam Altman’s departure. [9]
The 29 November announcement confirmed Altman’s return and Murati’s resumption of the CTO role. [10]
She left OpenAI in September 2024. In a subsequent interview, she explained that she was exploring her next professional chapter. [11]
Thinking Machines Lab: research and products
Dartmouth dates the founding of Thinking Machines to December 2024. [1]
The company aims to develop AI that people can understand, customise and use collaboratively. Its approach combines multimodal models, training infrastructure and joint research-and-product design, intending to learn from real uses and publish some of its technical work. [3]
In July 2025 it announced a $2 billion funding round led by a16z. Wilson Sonsini, which advised on the transaction, documented participation by investors including NVIDIA, Accel, ServiceNow, Cisco and AMD. The capital supports a company separate from OpenAI, led by Murati. [15]
Tinker: adapting models to specialist knowledge
Thinking Machines introduced Tinker on 1 October 2025. The platform enables model fine-tuning through an API, giving researchers and developers control over algorithms and data while the service manages distributed training, resources and failure recovery. It uses LoRA adaptations to share computing capacity between training runs. [16]
In its October 2026 review, the company described expanded models and modalities, role-based access controls, audit logs and tools for tracking sessions and costs. This evolution connects specialist research with the operational needs of organisations training their own models. [17]
A Bridgewater AIA Labs study with Thinking Machines, published in June 2026, used Tinker to fine-tune a model for classifying and filtering financial documents. Its authors evaluated agreement with expert criteria on specific tasks; they did not demonstrate investment returns or a general replacement for professional judgement. [20]
Inkling and open models
In July 2026 Thinking Machines released Inkling with weights available for download and fine-tuning. It presented the model as an adaptable multimodal foundation working with text, images and audio. The company sought to combine capabilities and customisation, without claiming it was the strongest model on every task. [18]
Inkling 2 Small was announced on 5 November 2026. It was initially available through Tinker with closed weights. An Apache 2.0 release was planned after external safety evaluations: the announcement distinguishes the available product from a weight release that was still prospective. [19]
Infrastructure and NVIDIA partnership
In March 2026 Thinking Machines and NVIDIA announced a multiyear partnership to deploy at least one gigawatt of Vera Rubin systems, targeted to begin in 2027. It included NVIDIA investment and joint work on training and serving models. This is committed, planned capacity rather than an installation already established as operational. [4]
AI safety and international participation
Murati served on the United Nations High-level Advisory Body on AI. Its institutional biography documents her responsibility for research, product and safety during her OpenAI tenure. She brought the perspective of someone who had led development and deployment of generative systems. [12]
The body brought together 39 members from 33 countries, serving in a personal capacity for a twelve-month mandate beginning in October 2023. Its role was to analyse and propose international governance mechanisms. Membership belonged to this time-limited process, rather than a permanent UN position. [13]
In August 2026 Thinking Machines announced grants of up to $50,000 in Tinker credits for open-weight model safety research. Topics included safeguards’ resistance to subsequent fine-tuning, alignment failures and risk estimation. These are research questions and programmes, rather than a guarantee that harmful uses are absent. [21]
Creative-work debate and academic recognition
At Dartmouth in June 2024, she supported regulation of high-risk uses and acknowledged uncertainty about jobs lost, changed or created. She also discussed creator compensation and iterative development with users. [14]
Dartmouth included her among its 2024 honorary doctorate recipients, recognising her technology career. The distinction is honorary and separate from the Bachelor of Engineering she earned as a student at the institution. [22]
