Thinking Machines Lab, an AI startup founded by former OpenAI Chief Technology Officer Mira Murati, has released its first general-purpose AI model. The model's architecture directly incorporates technical achievements from Chinese AI laboratories.

According to a report, the foundational model, named Inkling, references the architecture of DeepSeek's DeepSeek-V3. It was further optimized using data generated by Moonshot AI's Kimi K2.5 model in a post-training phase. Inkling will be released as an open-weight model, allowing users to deploy it on their own servers and fine-tune it for specific applications.

This release comes as Chinese AI models are accelerating their global market presence. Data from the platform OpenRouter indicates that the overall usage of Chinese AI models has surpassed that of their U.S. counterparts this year.

Inkling's Positioning: Open-Weight and Customizable

Based on benchmark data shared by Thinking Machines, Inkling's overall performance currently trails behind Anthropic's Claude and OpenAI's flagship products, as well as some leading Chinese models. However, the company has chosen to release it in an open-weight format, differentiating it from closed-source products like ChatGPT and Claude.

Open-weight release means enterprises and developers can deploy the model on their own infrastructure and deeply customize or fine-tune it for specific business needs. In a blog post, Thinking Machines stated that most current AI tools are "trained in a few places and then frozen," rather than being shaped by users. The company argued that "to scale human will and judgment, we need AI that is as diverse and distributed as people themselves."

The company has also launched the Tinker platform, aimed at enterprise clients, allowing them to fine-tune and customize large language models for specific business applications. Notably, competitors like OpenAI also offer customizable open-weight models, and cloud platforms such as Amazon and Microsoft provide model fine-tuning services based on users' own datasets, indicating significant competitive pressure for Thinking Machines in this space.

Technical Approach: Leveraging the Chinese AI Ecosystem

The model's reliance on Chinese AI innovations is one of the most notable details of this release. Since its launch earlier this year, DeepSeek-V3 has sparked widespread discussion in the global AI community for its low cost and high performance, with its architecture becoming a reference for several emerging models. Kimi K2.5, a reasoning-enhanced model from Moonshot AI, was used in Inkling's post-training optimization phase.

Thinking Machines stated that Inkling is a model "trained from scratch" and provides full weights for user access. "Today, we advance our mission by releasing a model we trained from the ground up, with full weights, for people to make their own," the company said.

Market interest in Thinking Machines is largely driven by Murati's unique standing in the AI industry. She led the development of several core OpenAI products, including ChatGPT, the image generator DALL-E, and voice mode features. She also played a significant role during the brief ouster of CEO Sam Altman by the OpenAI board in November 2023. Murati left OpenAI in September 2024 and founded Thinking Machines in February of last year.

Approximately one year after its founding, the company has completed a $2 billion seed funding round, achieving a post-money valuation of $12 billion. Investors include venture capital firm Andreessen Horowitz, chipmakers NVIDIA and AMD, and hedge fund Jane Street.

Beyond the Inkling foundational model, Thinking Machines is actively expanding its product line to build a competitive edge in differentiated areas. In May, the company previewed its first "interactive model," which can perform actions based on user audio and video inputs, not just text comprehension, signaling a strategic move towards multimodal AI and integration into daily life scenarios.