AI Product Updates: New Features and Releases

JEV is a multiple-choice-only model that answered a browser agent's flight search in 7 seconds for $0.0039, and Vercel says it is the fastest-adopted model in its gateway's history, free there until September 25.

Qwen-Image-2.1 is a 7B open-weight model for image generation and editing with native RGBA, up to 10 reference images and precise portrait and product editing.

Virtuals launched Occupy on Base, where tokens build shared stock treasuries and holders elect AI Senates to direct the capital, and went live on Arc so builders can tokenize agents.

Qwen announced that ComfyUI supports Qwen-Image-2.1, so the new image model can run directly in node workflows.

vLLM announced day-0 support for Qwen-Image-2.1, letting generation, editing and transparent output be deployed as a single served model.

A board collects tasks from different agents such as Codex and Claude, showing running, done and failed status, with automatic summaries.

The Chinese-language version of the same pitch: a manager opens the board to see every agent's task status instead of asking each person.

Alibaba's Tongyi released Qwen-Image-2.1, which it says improves on the previous generation in text layout and portrait aesthetics, with several text rendering examples.

Users can circle several regions and have the model remove a watch, change hair color and swap clothing in all of them at once.

The new model balances several editing tasks, such as generating a full storyboard from a three-view character reference.

The new model can take up to 10 reference images at once and still produce high-quality results, according to the post.

The model now supports generating transparent-background images natively and can also edit transparent images directly.