Thinking Machines Lab, the AI startup founded by former OpenAI CTO Mira Murati, announced its first in-house AI model, Inkling. The model comes as a response to the growing demand for open-weight AI systems, particularly for developers and enterprises looking for greater control over how large language models are trained, adapted and deployed.
Unlike closed AI models like ChatGPT and Claude, the weights of Inkling's model are available for outside developers to download, modify, and fine-tune. ChatGPT from OpenAI and Claude from Anthropic are mainly accessed through hosted products or APIs, where users can interact with the model but cannot directly change the underlying weights.
According to Thinking Machines, Inkling is a generalist model that was developed for agentic tasks, coding, reasoning and structured outputs. The company said, “That breadth matters for customization and real-world use: different users need models that can adapt to very different workflows, not just excel on benchmarks.”
Inkling is a 975 billion-parameter LLM with a one-million-token context window. However, it does not use all 975 billion parameters in each of its tasks. Given its mixture-of-experts architecture, it activates around 41 billion parameters for a given task, making the model more efficient.
The company's blog said Inkling was trained from scratch on 45 trillion tokens of text, image, audio and video. The model now generates textual outputs such as code, structured data and formatted/diagrammatic artifacts.
The model can also be configured to adapt and adjust its “thinking effort,” allowing it to trade off speed and reasoning depth depending on the use case, according to Thinking Machine Labs.
The primary distinction is between openness and customization. Closed frontier models like ChatGPT and Claude are accessed through controlled apps or APIs, but must be used in a controlled environment. Inkling, by contrast, is an ‘open weight’ basic model, which enterprises can ‘tune’ using Thinking Machines' model customisation platform, Tinker,
This makes Inkling more attractive for organizations that would like to create domain-specific artificial intelligence systems with their own processes and information. However, it also makes companies more responsible for safety, alignment, and security following fine-tuning of the model.
Also Read: NVIDIA Invests in Thinking Machines Lab, Plans 1GW AI Chip Deployment Worth Billions
Thinking Machines noted that the model works well for coding and reasoning but acknowledged that it's ‘not the strongest overall model available today, open or closed.’
Instead, the company is positioning Inkling as the first step in a broader model family. Alongside the main model, it also showcased Inkling-Small, a lighter model with only 12 billion active parameters that has lower costs and faster response time.