{"id":22951,"date":"2026-07-30T00:00:00","date_gmt":"2026-07-30T00:00:00","guid":{"rendered":"https:\/\/scannn.com\/introducing-inkling-small-thinking-machines-lab\/"},"modified":"2026-07-30T00:00:00","modified_gmt":"2026-07-30T00:00:00","slug":"introducing-inkling-small-thinking-machines-lab","status":"publish","type":"post","link":"https:\/\/scannn.com\/lv\/introducing-inkling-small-thinking-machines-lab\/","title":{"rendered":"Introducing Inkling-Small - Thinking Machines Lab"},"content":{"rendered":"\n
\n

Today, we are releasing Inkling-Small, an efficient open-weights model that achieves comparable performance to Inkling<\/a> at a quarter of its size.<\/p>\n

Inkling-Small is a Mixture-of-Experts transformer with 276B total parameters, 12B active, trained on NVIDIA GB300 NVL72 systems. Like Inkling, it features native reasoning over audio and images, variable thinking effort, a context window of up to 1M tokens, and well-rounded performance across a range of benchmarks.<\/p>\n

Parameter comparison between Inkling-Small and Inkling.<\/figcaption><\/figure>\n

Compared to Inkling, Inkling-Small achieves comparable performance with much less compute. Across agentic tool use (Terminal-Bench 2.1), reasoning (HLE text-only, no tools), and instruction following (IFBench) benchmarks, Inkling-Small is more efficient than Inkling and competitive with other models in its weight class. Furthermore, its variable thinking effort<\/a> lets users easily adapt it to target their use case, balancing cost and performance.<\/p>\n

\n
\n
\n
\n

\n Inkling-Small 12B active<\/span> (effort sweep)<\/span><\/span>
\n Inkling 41B active<\/span> (effort sweep)<\/span><\/span>
\n Comparison models<\/span>\n <\/p>\n<\/div><\/div>\n