A local model like Inkling 2 Mini does not make an AI assistant free for a small business. It moves the cost from monthly per-token API bills to one-time hardware and setup, and most of the budget still goes to integration work around the model.
Hugging Face introduced Inkling 2 Mini from Thinking Machines Lab. It is a multimodal Mixture-of-Experts model with about 26B total and 3B active parameters, a 1M-token context, and image and text input. Its middle layers run in a loop three times, so it behaves like a bigger model while staying small enough to run locally through llama.cpp or MLX.
What this means for a small business
Document flows are the clearest fit: invoices, contracts and scanned receipts processed on your own machine, with no per-page fees and no client data leaving the office. The model is only part of the cost, though. Someone still has to connect it to email, the CRM and storage, then test it on real files. That work is what drives the cost in how I break down project costs. The quality claims also need checking. The authors call their comparison "vibe tests" and say more validation is needed. The model also trails Qwen3.8-27B on AutomationBench, the benchmark closest to workflow automation.
Why it matters: If API bills or data residency are holding back your AI rollout, run Inkling 2 Mini on a test machine first. Compare its output with a cloud model on the same week of real documents.
Source: Hugging Face