An artificial intelligence startup called Thinking Machines made waves Wednesday with the unveiling of a new AI model that could offer users a rare Western-built alternative to the open-source AI tools currently dominated by Chinese technology companies.
The new model, called Inkling, falls into the category of open-weight AI — meaning anyone can download it, run it on their own systems, and modify it to fit their needs. This sets it apart from proprietary models that keep their underlying code locked away from users.
Inkling marks the first general-purpose AI model to come from Thinking Machines, a San Francisco company founded last year by Mira Murati, who previously served as chief technology officer at OpenAI.
The company had previously launched a product called Tinker back in October, which is designed to help users customize AI models. Inkling is now accessible through Tinker as well as other platforms used by developers.
In terms of raw size, Inkling is a heavyweight — boasting 975 billion parameters. Parameters are the internal variables that shape how an AI system understands and responds to information, and this figure places Inkling among the largest models of its type.
The release comes at a time when Western open-source AI development has fallen behind China’s. That gap widened after Meta shifted away from its open approach following the underwhelming reception of its Llama 4 model last year, leaving a void in the market.
With fewer Western open-source options available, many businesses have turned to Chinese-developed models as cost-effective alternatives to expensive closed-source products. Hedge fund Bridgewater Associates, for instance, used Thinking Machines’ Tinker platform to build a customized version of Qwen — a model created by China’s Alibaba — which the firm said outperformed leading proprietary models while costing less.
To demonstrate Inkling’s capabilities, Thinking Machines released a set of benchmarks comparing its performance against closed models from Anthropic, Google, and OpenAI, along with top open-source models, most of which come from Chinese AI labs.
While those competing models still hold an overall performance advantage, Inkling showed strong results — particularly when it came to agent-related tasks — which could attract attention from businesses and developers looking for alternatives.
