Thinking Machines: Inkling (free)

Thinking Machines Inkling is described as a multimodal mixture-of-experts model created by Thinking Machines Lab. Its architecture activates 41 billion parameters from a total of 975 billion, combining a broad expert pool with selective computation intended to support demanding general-purpose workloads.

The model is positioned for reasoning, coding, agentic systems, and tool-based applications. Its multimodal input and text-output design also make it a practical candidate for workflows that combine documents, images, or audio with analysis, code generation, and instructions for interacting with external tools. No supplied evidence establishes particular benchmark advantages or training-scale gains, so its strongest supportable fit is architectural and task-oriented rather than performance-leading.