The most capable open AI model of the year just became a free download — and it can hear, see, and read your operation's data without any of it leaving your building.
Thinking Machines released Inkling, their first open-weights model, under Apache 2.0: text, images and audio handled natively in one architecture, 975B parameters with 41B active per task, a 1-million-token context window, and reasoning effort you can dial up or down to trade speed for accuracy. Weights are on Hugging Face, fine-tuning is available on Tinker, and a lighter Inkling-Small is previewed.
If data privacy is what's kept AI out of your operation, open-weights releases like this are the green light to pilot in-house.
Models of this class make private, on-premise AI over PODs, gate-call recordings and control-tower data realistic: audio understanding for driver check-in calls, vision for damage documentation, and long context for entire tender documents — with no data leaving your infrastructure and no per-token bills.
https://thinkingmachines.ai/news/introducing-inkling/