Thinking Machines launched open-weight Inkling-Small
Thinking Machines is releasing Inkling-Small’s full weights while claiming near-Inkling performance at a quarter of the size.
The model has 276 billion total parameters, 12 billion active, and a context window up to one million tokens. It is available on Hugging Face, with fine-tuning through Tinker and multimodal chat in Tinker Playground. Thinking Machines says it uses less compute than Inkling and reports stronger results on some reasoning and coding benchmarks, while Inkling remains better on knowledge coverage and factuality. Safety testing follows the Inkling process, with internal testing and external red-teaming cited. TestingCatalog's note
The model has 276 billion total parameters, 12 billion active, and a context window up to one million tokens. It is available on Hugging Face, with fine-tuning through Tinker and multimodal chat in Tinker Playground. Thinking Machines says it uses less compute than Inkling and reports stronger results on some reasoning and coding benchmarks, while Inkling remains better on knowledge coverage and factuality. Safety testing follows the Inkling process, with internal testing and external red-teaming cited. TestingCatalog's note
score 8