Thinking Machines Unveils Inkling Model with Multimodal Reasoning.

Thinking Machines has announced the release of its new AI model, Inkling, which comes with its full weights openly available. This model is designed with a Mixture-of-Experts architecture, boasting 975 billion total parameters with 41 billion active, and offers native reasoning capabilities across text, images, and audio data. A key feature is its expansive context window, supporting up to one million tokens, alongside a controllable reasoning effort mechanism. Inkling is engineered for efficiency, promising lower token usage while maintaining comparable performance. It supports fine-tuning from its launch day via the Tinker platform and is noted for its robust agentic coding and tool-use functionalities. The model is compatible with major inference platforms, and a smaller version, Inkling-Small, is anticipated in the future, all trained from scratch by Thinking Machines.