Black Forest Labs releases FLUX.2

Background

Black Forest Labs (BFL) develops the FLUX line of image models. For FLUX.2, the company paired an image generator with a vision-language model from Mistral AI, a model that reads images and text together.^[1]

What happened

On 25 November 2025 Black Forest Labs released the FLUX.2 image models.^[1] The family includes hosted [pro] and [flex] models, and the 32B open-weight [dev] model.^[1] FLUX.2 couples a Mistral-3 24B vision-language model with a rectified flow transformer.^[1]

Key details

The models support output up to 4 megapixels and up to 10 reference images.^[1] The FLUX.2 VAE was released under the Apache 2.0 license, a permissive open-source license that allows reuse, including in commercial work.^[1]

Follow-up

The smaller FLUX.2 [klein] models followed on 15 January 2026.^[2]

Impact

With the release, hosted and open-weight options were offered within one family, so users could choose between calling a service and downloading a model.^[1]

Some of the terms in this release may be unfamiliar. A vision-language model is a neural network that accepts images and text together and produces output about them. A rectified flow transformer is an image generator that learns to move from random noise toward a finished picture. Combining the two means the system can take written instructions and visual inputs into account at the same time.

Hosted and open-weight variants serve different needs. Hosted models are reached through a service, usually with charges based on use, while open-weight models can be downloaded and run by others within the license that applies to each one. The FLUX.2 release offered both kinds within one family, so developers and everyday users could choose the route that suited them.^[1]

Apache 2.0 is a permissive open-source license. Releases under it generally allow reuse, including commercial use, and the FLUX.2 VAE was released under it, while FLUX.2 [dev] required a separate license for commercial use.^[1]