Five AI Labs Ship New Models, From FLUX 3 Image Klein to OmniHuman 2.0

Black Forest Labs, StepFun, Ant Group, Krea and BytePlus each released or previewed new AI models on October 15, 2026, according to aitimes.com. The releases span image generation, large language models and talking-head video.

Black Forest Labs released the weights of FLUX 3 Image Klein, an image generation and editing model with 5.6B parameters. It supports up to 10 reference images, 4K image generation, region-selective editing and element placement. The company said on X that it runs on a single consumer GPU. FLUX 3 Image Klein is free for commercial use by individuals and companies with annual revenue under $5 million; larger firms need a separate contract or the API.

StepFun released the weights of Step 5 Preview, a mixture-of-experts model with 600B total parameters and 27B active per token, supporting up to 1M tokens of context. It targets software development and professional knowledge work, with an emphasis on finance data analysis. StepFun had announced the model on X on September 20, 2026, calling it its new flagship for agentic work.

Ant Group released Ling-3.1-flash under the MIT license. It is a mixture-of-experts model with 560B total parameters and 25B active per token, and it also supports up to 1M tokens of context.

Krea opened the beta of Krea 3, which performs 2K image edits in under 8 seconds and uses up to 6 reference images. The model can be trained on specific products, people or styles to keep appearances consistent across images. Krea announced the beta on X on October 15.

BytePlus unveiled OmniHuman 2.0, which turns one image plus audio into a talking or singing video of up to 5 minutes. It generates expressions and gestures matching speech content, rhythm and intonation, and accepts text instructions to control motion and camera movement. BytePlus announced it on X on October 15.