Thinking Machines Lab released its first AI model, called Inkling, in July 2026. The open-weight model can be downloaded and modified by researchers and startups, and was trained from scratch to process audio, video, and text input.
Inkling contains 975 billion parameters, placing it among the larger publicly available models. Because of its size, it requires a cluster of specialized chips to run. The company says the model is capable of advanced reasoning and coding, though it does not top popular benchmarks. Notably, Thinking Machines used Inkling to fine-tune and improve itself during development.
During training, researchers observed an unusual behavior: Inkling dropped natural language explanations from its reasoning process, apparently treating grammar as unnecessary overhead. The company intervened and restored natural language reasoning to keep the model’s decisions more explainable, according to a source familiar with the process who was granted anonymity to speak freely.
Thinking Machines was founded in February 2025 by several former OpenAI executives and researchers, including Mira Murati, who served as CTO and briefly as CEO of OpenAI; John Schulman, an OpenAI cofounder who played a key role in developing ChatGPT; and Lilian Weng, a former VP who led safety and robotics work. The startup raised the largest seed funding round in history, valuing it at $12 billion at the outset.
The release fits with a position the company outlined in a recent blog post, arguing that AI should be decentralized rather than controlled by a small number of companies. Open-weight models are generally cheaper to run than closed models, which typically require paid access, and can be more easily adapted for specific tasks. Thinking Machines says Inkling’s performance is comparable to the leading open-weight models, which currently come from China.
The launch could help Thinking Machines compete with established players such as Anthropic and OpenAI. Anthropic, another company founded by OpenAI alumni, recently filed for an IPO that could value it at more than a trillion dollars.
Source: WIRED