FLUX 3 is Black Forest Labs’ multimodal frontier model that jointly learns from images, video, and audio within a unified architecture and can be extended to predict actions.

The rollout is phased: FLUX 3 Video and Action are available via API to select partners in early access, with image generation coming soon and a dev open-weight version for local use planned later in 2026.

BFL describes its approach as “Self-Flow,” enabling aligned understanding and generation of multimodal data across images, video, audio, and physical action.

Historically, Flux-style open-source versions competed for prominence, with open-weight Dev releases representing a strategic path for local use.

BFL emphasizes its open-weight heritage, notable partners and customers (Canva, Burda, Magnific, Krea, Picsart), and a substantial funding round valuing the company at about $3.25 billion.

BFL has not tied any crypto tokens or digital assets to FLUX, underscoring traditional venture funding over crypto-backed models.

Benchmark and pricing transparency are limited at launch; early data is preliminary and should be treated with caution due to the early-access nature.

There is no public API access, pricing, benchmarks, or downloadable weights at launch; faster open-weight versions are planned later in the year, with Dev offering an open-weight multimodal backbone.

In head-to-head tests, FLUX 3 output was preferred by human evaluators over competing models like Runway Gen-4.5 and Luma Ray 3.2, with competitive standings against Gemini Omni and Seedance.

Early benchmarks show FLUX 3 favored in about 77% of comparisons against Runway Gen-4.5 in initial evaluations.

Initial tests show FLUX 3 outperforming several rivals on short clips, though margins vary against stronger competitors like Kling v3 Pro and other versions.

The FLUX lineage began in 2024 with founders who contributed to Stable Diffusion, and Flux models have historically competed with MidJourney and Stability AI’s variants.