Thinking Machines Lab has moved from promise to product with the release of Inkling, its first artificial intelligence model. The company, started by exiles from OpenAI, is entering a crowded AI race with an open-weight model designed to work across text, audio and video.
The release matters because it shows how the startup wants to compete: not only by building powerful AI, but by making a model that researchers and startups can download, study and modify.
What Inkling is
Inkling is the first model released by Thinking Machines Lab. According to the company, it was trained from scratch to make sense of audio and video input as well as text.
The model is open-weight. In practical terms, that means outside researchers and startups can download it and modify it for their own work. That is a different route from closed models, which are typically accessed through paid services and are harder for outside teams to adapt.
Thinking Machines Lab says Inkling is not the top model on popular benchmarks. Even so, the company says it performs well across many tasks and is capable of advanced reasoning and coding.
Inkling is also large. The model has 975 billion parameters and needs to run on a cluster of specialized chips. That makes it open to modification, but not necessarily simple for every user to operate on ordinary hardware.
Why open-weight AI matters here
The open-weight release fits the broader direction Thinking Machines Lab has described for AI. In a recent blog post, the company said the technology should not be controlled by only a few companies. It argued for a more decentralized approach, where more people can build their own models using their own data.
That position is important because open-source and open-weight models have become popular for concrete reasons. They can be cheaper to run than closed models. They can also be changed more easily for different tasks.
For startups and researchers, that flexibility can matter as much as raw benchmark standing. A model that can be modified gives builders more room to adapt the system to their needs, test ideas and work with their own data.
The source article notes that the best open-weight models currently come from China. Thinking Machines Lab says Inkling offers a level of performance similar to those models. That claim places Inkling directly inside a competitive part of the AI market, where openness, cost and performance all matter.
The model helped improve itself
One of the more notable details in the release is how Thinking Machines Lab used Inkling during its own development. The lab used the model to fine-tune and improve itself, a sign of how AI models are increasingly being used to build AI.
The training process also revealed a change in how Inkling handled reasoning explanations. Like other models, Inkling usually produces a natural language explanation for complex reasoning. According to the company’s blog post, efficiency pressures changed that behavior over time.
“the chain of thought became more concise over time, dropping grammatical overhead while remaining comprehensible and leaving the final response unaffected.”
That observation is narrow but significant. It suggests the model’s reasoning explanation became shorter during training while still remaining understandable and without changing the final answer, according to the company.
For users focused on reasoning and coding, this detail helps explain why the company is highlighting more than benchmark scores. Thinking Machines Lab is presenting Inkling as a capable system with a training story that reflects current trends in model development.
Who is behind Thinking Machines Lab
Thinking Machines Lab was founded in February 2025 by several executives and researchers from OpenAI. Its founders include Mira Murati, who served as CTO and briefly CEO of OpenAI; John Schulman, a cofounder of OpenAI who played a key role in developing ChatGPT; and Lilian Weng, a former VP at OpenAI who led work on safety and robotics.
The startup received the largest seed funding round in history, valuing it at $12 billion out of the gate. Before releasing Inkling, the company released Tinker, a tool for fine-tuning models. It also showcased a tool for natural voice interactions and published machine-learning research.
That background gives the Inkling release extra attention. The company is not arriving as an unknown lab with a single model. It already had funding, prominent leadership and earlier work around fine-tuning, voice interaction and research.
How Inkling changes the competitive picture
OpenAI helped kick-start the AI boom with ChatGPT, but companies led by former OpenAI talent have become important competitors. The source article points to Thinking Machines Lab and Anthropic as examples of defector-led companies that have muscled into the space.
Anthropic recently filed for an IPO, which could value the company at more than a trillion dollars. Its model Claude has become popular with many businesses, especially for coding skills.
Thinking Machines Lab is now trying to establish itself as a legitimate player in that same fast-moving, high-spending AI race. Inkling gives the company a concrete model release to point to, rather than only tools, research and leadership credentials.
The release does not present Inkling as the best model on every benchmark. Its importance is different: it gives Thinking Machines Lab an open-weight foundation model with multimodal input, reasoning and coding capabilities, and a stated philosophy of more decentralized AI development.
That combination makes Inkling a strategic first model. It is a technical release, but also a signal about where Thinking Machines Lab wants to stand in the debate over who gets to build, adapt and control advanced AI systems.