Thinking Machines has released the Inkling 2 Small model

Startup Thinking Machines, founded by former OpenAI CTO Mira Murati, has introduced the Inkling 2 Small language model. The new model is built on a Mixture-of-Experts architecture and contains 594 billion parameters.

According to the developers, the model holds a leading position among Western open-weight systems in terms of generation quality. The architecture is focused on performing complex autonomous tasks and increased agentic performance in practice.

Improving autonomous capabilities required significant hardware resources. Using the model is associated with a sharp increase in token consumption when processing requests.

The release of the system expands the segment of large open-weight models for enterprise tasks and research labs that require control over their infrastructure.