Yes—but “rival to OpenAI” is shorthand, not the company’s formal launch description. Former OpenAI chief technology officer Mira Murati publicly unveiled Thinking Machines Lab on February 18, 2025, with a prominent research team and a mission centered on customizable, understandable and collaborative AI. It did not launch a ChatGPT replacement that day. By August 18, 2026, the company had added Tinker, a managed fine-tuning platform, and Inkling, an open-weights model, while pursuing interactive AI and large-scale computing infrastructure.
What happened on February 18, 2025?
Murati announced that she had co-founded and was leading Thinking Machines Lab as chief executive. The company emerged from stealth through a public statement and company launch post, but disclosed a research and product direction rather than a ready-to-use consumer application. TechCrunch’s launch coverage reported the company’s mission and leadership.
The announcement did not include a named flagship model, public API, chatbot, pricing, performance benchmarks, product timetable or company-confirmed funding amount. Axios reported that the startup did not provide a first-product timeline or detailed product specifications.
Why Mira Murati matters
Murati joined OpenAI in 2018 and became its CTO in 2022. In that senior technical and product role, she was closely associated with OpenAI’s work on ChatGPT, DALL-E and Codex. She briefly served as interim CEO during OpenAI’s November 2023 leadership crisis and left the company in 2024 before starting Thinking Machines Lab.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
#1 Best Overall
That history gives the new lab immediate credibility and explains why coverage often calls it an OpenAI competitor. It does not mean Murati personally created ChatGPT or OpenAI’s other products; those were produced by large teams.
Who joined the founding team?
- Mira Murati: co-founder and CEO.
- John Schulman: chief scientist and OpenAI co-founder.
- Barret Zoph: CTO and former OpenAI research executive.
Launch-era reporting described roughly 30 researchers and engineers recruited from OpenAI, Meta, Mistral, Google DeepMind, Character AI and other laboratories. That was an early team description, not a current headcount; the reported figure should not be treated as a permanent workforce total. TechCrunch later reported Bob McGrew and Alec Radford joining as advisers in April 2025.
What Thinking Machines Lab says it is building
The launch mission had three connected aims: help people adapt AI to their own needs, build foundations for more capable systems, and grow a community around AI research and understanding. The technical themes were:
Rank #2
- Multimodal systems that work across text, images and other inputs.
- Human-AI collaboration rather than one-size-fits-all automation.
- Models that can be customized for individuals, organizations and specialized workflows.
- Frontier capabilities in areas such as science and programming.
- Research intended to make AI behavior more understandable, with safety work and selective sharing of code, datasets, specifications and practices.
The emphasis is therefore not simply “build a bigger chatbot.” The company’s thesis is that useful AI should be interactive and adaptable while preserving human judgment. Its later statements continue that direction, including the company’s human-centered position.
Is it really an OpenAI rival?
It is reasonable to call Thinking Machines Lab a potential OpenAI competitor, but misleading to describe it as a ChatGPT substitute launched in February 2025.
| Question | What the evidence supports |
|---|---|
| Industry competition | Yes. It has OpenAI alumni, frontier-model ambitions, major financing reports and a large future compute plan. |
| Consumer product at launch | No. The February 2025 announcement named no chatbot, model, API, price or release date. |
| Strategic emphasis | Thinking Machines publicly stresses customization, interaction, research infrastructure and open-weight releases; OpenAI is best known for ChatGPT, hosted models and APIs. |
| Direct product replacement | Not established. Tinker is training infrastructure, not a general-purpose assistant. |
The contrast is directional, not absolute. Thinking Machines’ NVIDIA announcement also refers to frontier-model training, so its open-model and customization work should not be read as evidence that it has abandoned frontier research.
What became available after the launch?
Tinker: a model-training platform
Thinking Machines announced Tinker on October 1, 2025. It is a managed API for fine-tuning open-weight language models while the service handles scheduling, resource allocation, distributed-training infrastructure and failure recovery. Its low-level operations include forward_backward, optim_step, sample and save_state. Tinker uses LoRA adapters, which update a smaller set of parameters and allow compute to be shared between training runs. The launch announcement initially described a private beta that was free to start, with usage pricing to follow.
General availability arrived on December 12, 2025, ending the waitlist. The release added fine-tuning support for Kimi K2 Thinking, an OpenAI-compatible sampling interface and vision input through Qwen3-VL. The current Tinker page lists models from Qwen, DeepSeek, Moonshot, NVIDIA, GPT-OSS and Thinking Machines itself; supported models and prices can change.
Who Tinker is—and is not—for
- Good fit: ML researchers, university labs, developers building specialized agents and organizations with proprietary training data.
- Required work: supply suitable data, define a training or reinforcement-learning objective, evaluate outputs and manage safety and behavior.
- Poor fit: casual users wanting a chatbot, teams without evaluation expertise, or buyers seeking a complete application rather than model-training infrastructure.
The current page describes usage-based pricing in U.S. dollars per million tokens and checkpoint storage at $0.10 per GB-month. Verify model-specific training and inference rates before committing, because the catalog and pricing are volatile. A managed service also means considering whether sending training data to an external provider fits privacy, compliance and data-residency requirements.
Inkling: an open-weights model
On July 15, 2026, the company announced Inkling as a generalist open-weights model with multimodality, agentic coding and tool use, adjustable thinking effort and customization through Tinker. The announcement also described safety and epistemic features. The company’s news page later listed Inkling-Small as introduced on July 30, 2026. Access, licensing and download terms should be checked on the Inkling announcement and current news page.
“Open weights” is narrower than “open source.” It does not, by itself, establish that the training data, full training code or every commercial use is open.
Timeline from startup to platform
| Date | Development |
|---|---|
| February 18, 2025 | Thinking Machines Lab publicly launches; Murati is CEO, Schulman chief scientist and Zoph CTO. |
| April 8, 2025 | TechCrunch reports Bob McGrew and Alec Radford joining as advisers. |
| June–July 2025 | TechCrunch reports a $2 billion seed round, with valuation reports of about $10 billion and later $12 billion. |
| October 1, 2025 | Tinker enters private beta. |
| October 29, 2025 | Research and teaching grants are announced, including $250 per student for teaching grants and research grants starting at $5,000. |
| December 12, 2025 | Tinker becomes generally available. |
| March 10, 2026 | NVIDIA partnership announces a planned one-gigawatt deployment of next-generation systems, targeted to begin in 2027. |
| May 19, 2026 | Interactivity research grants are announced, including grants of $100,000 plus Tinker credits. |
| July 15, 2026 | Inkling open-weights model is announced. |
| July 30, 2026 | The company’s news page lists Inkling-Small as introduced. |
Funding, valuation and compute
Thinking Machines did not disclose funding at the February 2025 launch. Earlier reporting said Murati was seeking more than $100 million, but that was not a confirmed launch figure. Later TechCrunch reports said the company closed a $2 billion seed round in 2025 and placed its valuation at approximately $10 billion, then approximately $12 billion. Those are media-reported estimates, not figures confirmed in the launch announcement; see the June report and July report.
Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Fix the driver behind crashes, sound loss and screen glitches3Repair Windows errors before they cause bigger problemsBest Value
On March 10, 2026, Thinking Machines and NVIDIA announced a multiyear partnership involving at least one gigawatt of next-generation NVIDIA Vera Rubin systems, with deployment targeted for early 2027. NVIDIA also made a significant investment, according to the company announcement. One gigawatt describes planned infrastructure capacity—not a consumer product or proof that an equivalent cluster is already operating.
What the strategy means for customers and investors
Tinker suggests a possible commercial model built around managed customization and training infrastructure rather than only a single proprietary chatbot. That is an inference from the product’s positioning, not a statement that Tinker is the company’s exclusive business.
Customization can improve a model’s usefulness for a defined workflow, but it creates its own risks:
- Private data may require stronger contractual, privacy and residency controls.
- Poor data or weak evaluations can make a customized model less reliable than a general model.
- Fine-tuned behavior can drift as data and objectives change.
- Open weights do not remove safety, misuse or licensing questions.
- A smaller specialized model is not automatically better than a larger frontier model.
For finance and business readers, the key question is execution: whether the lab can turn talent, capital and planned compute into tools that deliver measurable value for researchers, developers and enterprises. A large seed round or future gigawatt commitment is an input, not evidence of product-market fit.
What can people use today?
As of August 18, 2026, Tinker is the clearest practical entry point for technical users, while Inkling is the company’s public open-weights model line. Neither should be presented as a normal consumer replacement for ChatGPT. Prospective users should check the official Tinker page for current eligibility, model support, data handling and usage rates, and the Inkling materials for access and licensing details.
Bottom line
Thinking Machines Lab began on February 18, 2025 as a high-profile AI research startup led by a former OpenAI CTO and staffed by notable researchers. It was not a finished ChatGPT rival on launch day. By August 2026, it had become a broader bet on customizable and interactive AI: Tinker supplies managed fine-tuning infrastructure, Inkling supplies open weights, and the NVIDIA agreement supports longer-term frontier-scale ambitions. Whether that becomes a durable OpenAI competitor will depend on the usefulness, safety and economics of those products—not simply on Murati’s name, reported valuation or promised compute.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
