FLUX 3 (early access)
FLUX 3: one multimodal flow model for image, video with native audio, and robot-action prediction; open FLUX 3 Dev promised.
Jointly trained on images, video and audio with the Self-Flow approach; video up to 20 seconds with multilingual speech. Video launched 2026-08-04, FLUX 3 Action (7B world-action model) on 2026-09-23. Early access via API and private weights; an open-weight backbone is planned.
- Date
- Thursday, 23 July 2026
- Lab
- Black Forest Labs
- Kind
- model
- Access
- closed API
Figures
| Measure | Value | Measured by |
|---|---|---|
| Preference vs Runway Gen-4.5 / Luma Ray 3.2 / Grok Imagine Video | 77% / 93% / 69% BFL's own preliminary human comparisons | company |
Benchmarks are preliminary and company-run. The robotics partner named on the BFL blog is mimic. Later blog posts include FLUX Video Upscale 2K/4K on 2026-08-20.
Sources
- bfl.ai/blog/flux-3
- bfl.ai/blog
- www.marktechpost.com/2026/07/26/black-forest-labs-releases-flux-3-a-multimodal-flow-model-
This record was checked against its sources on 6 October 2026. How we check