Video generation has a waiting problem. Even good models can take minutes to render a few seconds of footage, which makes them awkward inside products that need an answer now. fal says its new model closes most of that gap. The company announced FLUX 3 Max on October 19, a video model post-trained by fal Research in collaboration with Black Forest Labs and built on the FLUX 3 foundation model.

The headline numbers come in two kinds, and it is worth keeping them separate. The quality figure is independent: in preliminary human preference evaluations on the Artificial Analysis Video Arena, FLUX 3 Max scored 1,139 Elo across 4,868 votes, placing it among the highest rated image-to-video models evaluated. That is a 50-point improvement on the original FLUX 3, which scored 1,089. The speed figures are fal’s own benchmarks. In fal’s latency tests, the model generates a 10-second video in 10.95 seconds, against 277.80 seconds for Seedance 2.5 at the same reported Elo rating. Against its own predecessor, generation time fell from 126.44 seconds to 10.95 seconds, an 11.5x improvement.

Abstract illustration of video frames generating at high speed
Illustration created for GPTOnline.

What the two teams say

The launch is structured as a partnership between a foundation model lab and an inference specialist, and both sides framed it that way. “We built FLUX 3 as a foundation for others to build on,” said Robin Rombach, co-founder and CEO of Black Forest Labs. “FLUX 3 Max is a great example of what that looks like.”

Burkay Gur, CEO of fal, pointed to the combination of model work and systems work: “We believe some of the most important advances in generative AI will come from optimizing models and the systems that run them together.” Kaan Oktay, fal’s head of research, added that model performance does not stop improving when base training ends, and said the team is continuing to post-train FLUX 3 Max beyond this first iteration.

Abstract illustration of two research teams combining their work into one model
Illustration created for GPTOnline.

Availability and the fine print

FLUX 3 Max is available from October 19 on fal’s platform, covering text-to-video, image-to-video and keyframes-to-video generation through one API. The company is offering 75% off for the first three weeks of the launch.

Three caveats apply. First, the speed comparisons come from fal’s own benchmark setup, and rival models may perform differently under other conditions. Second, the Artificial Analysis rating is described by both companies as preliminary, with confidence intervals overlapping among the top-ranked models, so the exact ordering at the top is not settled. Third, this is explicitly a first iteration: fal says further improvements to visual quality and generation speed are already underway. Even with those caveats, a 10-second clip rendered in about 11 seconds changes what developers can build, because video stops being a background job and becomes something closer to a real-time response.

Sources