Thinking Machines debuts Inkling Small open source AI model nearing performance of predecessor at about 1/4 size

Thinking Machines, led by former OpenAI CTO Mira Murati, released Inkling-Small, a 276B-parameter multimodal reasoning model under an Apache 2.0 license. The model approaches or beats the larger 975B-parameter Inkling on several benchmarks while using far fewer active parameters per token and supporting up to a one-million-token context window.

The release highlights a broader shift toward efficient frontier models that preserve high-end performance while lowering inference and deployment costs. For enterprises, Inkling-Small could make advanced open-weight AI more practical for teams with limited GPU capacity, especially if fine-tuning and API pricing remain competitive.