Inkling Small

Inkling Small is an open-weights multimodal language model developed by Thinking Machines Lab, the AI startup founded by former OpenAI CTO Mira Murati. Released in July 2026, it is designed as a highly efficient, compact alternative to the company's flagship Inkling model. The system is built on a sparse Mixture-of-Experts (MoE) architecture, featuring 276 billion total parameters with only 12 billion active per token, which enables it to maintain high performance while significantly reducing the computational footprint required for inference.

The model natively processes text, image, and audio inputs, allowing for a wide range of cross-modal reasoning tasks. A standout feature of the Inkling family is controllable thinking effort, a mechanism that lets developers adjust the model's internal reasoning depth. This allows for a flexible balance between rapid response times for simple queries and intensive, high-accuracy processing for complex logical problems, such as agentic coding or calibrated forecasting.

Technical Capabilities

Inkling Small supports an expansive context window of up to 1,000,000 tokens, making it capable of analyzing massive datasets, long-form documents, and extensive multimodal contexts. Despite its smaller size compared to the 975B-parameter flagship, it maintains competitive scores on benchmarks like SWE-bench and various reasoning indices. The model is released under a permissive Apache 2.0 license, intended to provide a robust open-weights foundation for enterprise and research applications.

Beyond standard chat functions, the model is optimized for agentic workflows and complex tool-use. It integrates with the company's Tinker platform, which facilitates fine-tuning and the creation of specialized versions. By combining multimodal flexibility with efficient sparse architecture, Inkling Small serves as a versatile tool for developers looking to build customized, high-performance AI systems without the overhead of trillion-parameter models.