What’s new: For developers building agents on phones and laptops. It uses Liquid AI's new tri-hybrid design and is part of a family that also has vision, audio and other models. Not meant for coding agents.

What’s new: For teams running large open models. inclusionAI says it scores 44 on the Artificial Analysis Intelligence Index, up from 41 for the earlier version.

What’s new: For anyone turning very long documents, up to hundreds of pages, into text. Baidu claims top results on long-document benchmarks.

What’s new: For teams running automated agents. Usage is billed to the organization, and its model policies and budgets apply, instead of being tied to one person's account.

What’s new: For people running models locally. The makers report SWE Verified scores of 62.8 for the 4B and 59.8 for the 1.7B.