Thinking Machines Lab, the AI startup led by former OpenAI CTO Mira Murati, released its first in-house model Wednesday morning. Called Inkling, it arrives with an open-weight design that allows outside developers to download and modify it directly — a deliberate departure from the flagship models offered by OpenAI, Anthropic, and Google. Main Developments Inkling is a mixture-of-experts system with 975 billion total parameters, though it activates only about 41 billion for any given task. That design keeps very large models faster and cheaper to run. The model was trained on 45 trillion tokens of text, image, audio, and video, reasoning natively across all four modalities, according to the company's own release materials. For now, Inkling's outputs are limited to text, including code, styled artifacts, and structured data. The company hasn't yet enabled multimodal output, keeping its initial release focused on text-based tasks. Read also: 4 reasons OnePlus is winding down US and Europe operations Background Murati founded Thinking Machines Lab after leaving OpenAI, where she had overseen the development of GPT-4 and other major models. The startup's first release signals a strategic bet against the industry trend toward closed, proprietary AI systems controlled by a few large companies. Why It Matters Open-weight models give developers and companies the freedom to customize, audit, and deploy AI without depending on a single provider's API or licensing terms. Inkling's mixture-of-experts architecture also demonstrates a path toward more efficient large-scale AI — using only a fraction of total parameters per task — which could lower compute costs and energy consumption for users. What's Next Thinking Machines Lab has not announced a timeline for multimodal output capabilities or additional model sizes. Developers can now download and experiment with Inkling, and the company's next moves will likely depend on community adoption and feedback.