Thinking Machines Lab is an American artificial intelligence research company founded in 2024 by Mira Murati, the former Chief Technology Officer of OpenAI. The company is headquartered in San Francisco. Thinking Machines launched publicly in February 2025 with approximately 30 researchers and engineers, released its first product, the Tinker fine-tuning platform, in October 2025, and in May 2026 unveiled the TML-Interaction-Small model as the first entry in its Interaction Models family, a real-time audio, video, and text interaction system positioned against OpenAI GPT Realtime and Google DeepMind Gemini Live. As of May 2026, the company is reported to be raising at a $50 billion valuation, an unusually fast trajectory from its $12 billion seed round in July 2025.
At a glance
- Founded: 2024 in San Francisco. Public launch February 2025.
- Paraform Talent Density: Rank #1, score 0.817 (May 2026). Source.
- Status: Private. Approximately 30 researchers and engineers as of public launch; team scaled through 2025 and 2026.
- Funding: $2 billion seed at a $12 billion valuation in July 2025, led by Andreessen Horowitz. Reports in November 2025 of a $5 billion follow-on at a $50 billion valuation under negotiation.
- CEO: Mira Murati (founder)
- Other notable leadership: John Schulman (Chief Scientist; OpenAI co-founder; previously at Anthropic), Lilian Weng (former OpenAI VP), Jared Kaplan (advisor; Chief Science Officer at Anthropic). Barret Zoph, co-founder and former Chief Technology Officer, departed in January 2026 to return to OpenAI.
- Open weights: None as of May 2026.
- Flagship products: Tinker (October 2025, fine-tuning platform). TML-Interaction-Small (May 2026, the first model in the Interaction Models family; real-time audio, video, and text; research preview).
Origins
Thinking Machines Lab was founded by Mira Murati after her departure from OpenAI in September 2024. Murati had been Chief Technology Officer at OpenAI from 2018 through 2024, briefly served as interim CEO during the November 2023 board crisis, and was widely identified as a central figure in OpenAI product strategy through the period of GPT-4 and ChatGPT scaling. Her departure was part of a broader 2024 cohort of senior OpenAI departures including Ilya Sutskever (founded SSI), Bob McGrew, and others.
The company was founded quietly in late 2024 and launched publicly in February 2025 with approximately 30 researchers and engineers already on staff. The founding team was unusually concentrated in former OpenAI senior staff. Barret Zoph, formerly OpenAI's Vice President of Research, joined as co-founder and Chief Technology Officer. Lilian Weng, formerly OpenAI's Vice President of Safety Systems, joined as a co-founder. John Schulman, an OpenAI co-founder who had moved to Anthropic in 2024, departed Anthropic to join Thinking Machines as Chief Scientist. Jared Kaplan, Anthropic's Chief Science Officer, joined as an advisor.
In July 2025, Thinking Machines raised $2 billion in its first reported funding round at a $12 billion valuation, led by Andreessen Horowitz with participation from NVIDIA, AMD, Cisco, and Jane Street. The round was described as the largest seed round in venture-capital history and reflected the unusual reputation density of the founding team.
In October 2025, Thinking Machines released its first product, Tinker, a developer-facing fine-tuning platform that lets users customize existing AI models for specific tasks without the cost and complexity of distributed training. In November 2025, Bloomberg reported that the company was raising an additional $5 billion at a $50 billion valuation. In March 2026, Thinking Machines announced a strategic partnership with NVIDIA involving an undisclosed investment and a multi-year agreement to deploy one gigawatt of Vera Rubin computing capacity. In April 2026, TechCrunch reported that Google had deepened ties with Thinking Machines through a new multibillion-dollar agreement.
In May 2026, Thinking Machines unveiled the Interaction Models family, the company's first in-house model release, with TML-Interaction-Small as the inaugural entry. The model is a real-time audio, video, and text system designed for sustained human-AI collaboration sessions, with 276 billion total parameters in a mixture-of-experts architecture (12 billion active per token), 200-millisecond micro-turn chunking, and a reported 0.40-second turn-taking latency. Thinking Machines framed the release around what the company described as the "collaboration bottleneck": current interfaces force users out of the conversational loop through slow turn-based design, where natural collaboration requires concurrent input and output streams across modalities. On the company's internally-developed FD-bench V1.5 interaction-quality benchmark, TML-Interaction-Small reported a score of 77.8 against published comparison settings for OpenAI GPT-4 Realtime, Google DeepMind Gemini 3.1 Flash Live, and Alibaba Qwen 3.5 Omni-plus-realtime in the 45.5 to 54.3 range. The release was structured as a limited research preview with wider availability planned later in 2026; no API or public demo was published at launch.
Mission and strategy
Thinking Machines Lab's stated mission is to build AI that is "more useful, more knowable, and more reliable" through research that emphasizes scientific understanding alongside capability. Murati has framed the company's positioning as a contrast to product-velocity-first labs, emphasizing depth of research over rapid release cadence.
The strategy combines three threads. First, in-house frontier-model research, with the explicit plan to release Thinking Machines' own models in 2026 according to public statements from John Schulman. Second, developer-facing tooling that lets users adapt and fine-tune existing models, of which Tinker is the first product. Third, deep compute partnerships including NVIDIA's gigawatt Vera Rubin commitment in March 2026 and the deepened Google relationship reported in April 2026, which provides Thinking Machines with diversified frontier-scale training infrastructure across multiple silicon vendors.
The competitive premise is that the founding team's reputation, the depth of senior research talent, and the differentiated funding scale produce a research environment that can compete with OpenAI, Anthropic, and Google DeepMind on capability, while the developer-tooling and customization positioning provides a distinct commercial path that does not require beating the closed-weights flagship labs on raw benchmark results.
Models and products
- Tinker. Developer-facing fine-tuning platform launched October 2025. Lets developers customize existing AI models for specific tasks without the engineering complexity of distributed training. The first product the company released.
- Interaction Models family. Announced May 2026. The first in-house model line from Thinking Machines, positioned around real-time multimodal human-AI collaboration rather than turn-based chat. The inaugural entry, TML-Interaction-Small, is a 276 billion parameter mixture-of-experts model (12 billion active) handling audio, video, and text concurrently with time-aligned micro-turn chunking. Available as a limited research preview at launch with wider availability planned later in 2026.
The company's distribution strategy beyond Tinker and the Interaction Models research preview has not been disclosed.
Benchmarks and standing
The May 2026 TML-Interaction-Small release produced the company's first public benchmark result. On the FD-bench V1.5 interaction-quality benchmark introduced alongside the model, TML-Interaction-Small reported a score of 77.8, against published comparison settings for OpenAI GPT-4 Realtime, Google DeepMind Gemini 3.1 Flash Live, and Alibaba Qwen 3.5 Omni-plus-realtime in the 45.5 to 54.3 range. The benchmark is internally developed by Thinking Machines and external reproductions have not yet been published; the reported delta is consistent with the company's stated framing that real-time turn-taking, simultaneous speech, and concurrent multimodal input are the differentiators rather than raw single-modality capability.
No benchmark results have been published for Tinker (a tooling product rather than a model) or for any Thinking Machines research outside the Interaction Models family. The company's broader standing in the industry continues to rest on the founding team's reputation, the funding scale, and the strategic-investor partnerships. Industry coverage has placed Thinking Machines among the most-watched AI companies, and the reported $50 billion valuation negotiation in late 2025 indicates investor conviction in the team's eventual product output.
Leadership
As of April 2026, Thinking Machines Lab's senior leadership includes:
- Mira Murati, founder and Chief Executive Officer. Former Chief Technology Officer of OpenAI (2018 through 2024). Public face for the company on capability claims, fundraising, and partner relationships.
- John Schulman, Chief Scientist. OpenAI co-founder and senior research leader on reinforcement learning. Joined Anthropic in 2024 after leaving OpenAI, then departed Anthropic to join Thinking Machines as Chief Scientist. Has stated publicly that Thinking Machines will release its own models in 2026.
- Barret Zoph (departed January 14, 2026). Co-founder and Chief Technology Officer through January 2026; former Vice President of Research at OpenAI. Returned to OpenAI in January 2026 alongside Luke Metz and Sam Schoenholz.
- Lilian Weng, co-founder. Former Vice President of Research, Safety Systems at OpenAI.
- Jared Kaplan, advisor. Co-founder and Chief Science Officer of Anthropic; his advisory role at Thinking Machines is unusual given his concurrent senior position at Anthropic.
The company has reportedly hired aggressively from senior researcher and engineer ranks at OpenAI, Anthropic, and Meta AI / FAIR, as well as from academia. The reported team size as of public launch was approximately 30 researchers and engineers; it has scaled materially through 2025 and 2026 without public disclosure of current headcount.
Funding and backers
Thinking Machines Lab's funding history through April 2026 includes a single closed round and one round reported in negotiation but not yet closed:
- Seed (July 2025): Approximately $2 billion at a $12 billion post-money valuation, led by Andreessen Horowitz with NVIDIA, Accel, ServiceNow, Cisco, AMD, Jane Street, and a $10 million strategic investment from the government of Albania (founder Mira Murati's country of origin, requiring an amendment to Albania's 2025 budget). The round was the largest seed round in venture-capital history at the time.
- Reported follow-on (November 2025, not closed as of April 2026): Bloomberg reported that the company was in negotiations for an additional round at approximately a $50 billion valuation. The round structure and amount have not been formally announced; coverage suggested Murati was attracting former OpenAI colleagues including John Schulman, Barret Zoph, and Luke Metz.
Cumulative closed funding stands at approximately $2 billion as of April 2026.
The March 2026 NVIDIA partnership combined an undisclosed equity investment with a multi-year agreement to deploy one gigawatt of NVIDIA Vera Rubin computing capacity. The April 2026 Google deal, reported by TechCrunch on April 22, 2026, deepened the relationship between Thinking Machines and Google with a new multibillion-dollar agreement, the terms of which have not been publicly disclosed.
The combination of NVIDIA and Google strategic relationships gives Thinking Machines diversified frontier-scale compute access across both NVIDIA silicon and Google TPUs, an unusual configuration for a single company at this stage.
Industry position
Thinking Machines Lab occupies a structurally distinctive position among Insurgent labs. The combination of the largest seed round in venture-capital history, the senior OpenAI-departure founding team, the rapid valuation acceleration from $12 billion to a reported $50 billion in negotiation, and the simultaneous strategic relationships with NVIDIA and Google produces a profile that is closer in shape to a Frontier lab than to typical Insurgent companies.
The strategic risks are substantial. The company has not yet released a model, the technical direction has not been disclosed, and the $50 billion-class valuation depends on capability claims that have not been demonstrated publicly. The Tinker product is a tooling offering rather than a model, and is not the basis for the company's valuation.
The strategic strengths are equally distinct. The founding team's research credentials are unmatched outside the established Frontier labs. Compute access is diversified across NVIDIA and Google. Capital is sufficient to support multi-year frontier-tier training programs without near-term commercial constraints. The company's positioning as a research-first lab with developer-tooling commercialization is differentiated from the consumer-product orientations of OpenAI and Anthropic.
The Jared Kaplan advisor relationship is unusual given his concurrent role as Anthropic's Chief Science Officer. The structural implications for both companies, including potential conflicts of interest in talent recruitment or research direction, have not been publicly addressed.
Competitive landscape
Thinking Machines Lab competes with several Frontier and Insurgent labs:
- OpenAI. The previous lab of most of Thinking Machines' senior staff. Direct competitor for senior AI talent and frontier capability.
- Anthropic. John Schulman's previous lab and Jared Kaplan's current employer. Direct competitor for safety-and-research-positioned AI talent.
- Google DeepMind. Frontier-research peer; Google is also a strategic partner via the April 2026 deepened relationship.
- Safe Superintelligence (SSI). Closest peer Insurgent. Both founded by senior 2024 OpenAI departures, both pursuing high-capital research-first strategies, both pre-product on in-house models.
- AMI (Advanced Machine Intelligence). Yann LeCun's lab, also a 2025-founded Insurgent at the senior end of the cohort.
- Other Insurgent labs. Reflection AI, Magic, Periodic Labs, and the broader 2024 to 2025 founder-exodus cohort compete for senior AI talent and compute allocations.
Outlook
Several open questions affect Thinking Machines Lab's trajectory in 2026 and 2027:
- Wider availability of the Interaction Models family beyond the research preview. Thinking Machines has indicated a broader release later in 2026; the pricing structure, distribution surfaces (API, embedded SDK, consumer product), and developer-access tier are unannounced.
- Successor models in the Interaction Models family. TML-Interaction-Small is positioned as the inaugural entry, implying a Medium or Large variant on the roadmap. The parameter-count scaling and the corresponding latency-versus-quality trade-offs are the central technical unknowns.
- External reproduction of the FD-bench V1.5 results. The benchmark is internally developed; independent evaluations by third parties (Artificial Analysis, academic researchers, or the OpenAI / Google DeepMind realtime teams) will indicate whether the reported gap holds against the most current GPT Realtime and Gemini Live releases.
- The reported $50 billion funding round structure, terms, and lead investor.
- The Google partnership terms revealed in the April 2026 multibillion-dollar deepening of ties, and whether the partnership materially shapes the Interaction Models distribution path.
- Tinker adoption metrics and the commercial trajectory of the fine-tuning platform alongside the new model line.
- Continued senior-talent recruitment, particularly from OpenAI, Anthropic, and Meta AI.
- The unusual Jared Kaplan dual-role situation between Thinking Machines and Anthropic, and any clarification of the relationship as both companies scale.
- Any acquisition interest from Frontier or Incumbent labs at the reported $50 billion valuation level.
Sources
- Thinking Machines Lab: Interaction Models. May 2026 announcement of the Interaction Models family and TML-Interaction-Small.
- VentureBeat: Thinking Machines shows off preview of near-realtime AI voice and video conversation. Industry coverage of the May 2026 Interaction Models preview.
- TechCrunch: Mira Murati's Thinking Machines Lab is worth $12B in seed round. July 2025 seed round.
- TechCrunch: Google deepens Thinking Machines Lab ties with new multi-billion-dollar deal. April 2026 Google partnership.
- Yahoo Finance: Thinking Machines seeks $50 billion valuation in funding talks. Bloomberg-sourced November 2025 valuation reporting.
- Built In: Inside Thinking Machines Lab, Mira Murati's New AI Startup. Reference profile.
- Thinking Machines Lab official site. Tinker product page and company information.
- Wikipedia: Thinking Machines Lab. Reference material.
