fal's FLUX 3 Max video model claims top-tier image-to-video quality in about 11 seconds a clip

fal, the generative media inference platform, launched FLUX 3 Max on 19 October. It is a video model post-trained by fal Research from Black Forest Labs’ FLUX 3 foundation model, with BFL as collaborator. fal’s model page describes it as “a post-trained variant of FLUX 3 Video, tuned for stronger prompt adherence and better visual quality while optimized for lower latency.”

What fal claims

• Quality: 1,139 Elo across 4,868 votes in preliminary Artificial Analysis Video Arena image-to-video voting, “with overlapping confidence intervals among the highest-ranked models.” That is up 50 points from the original FLUX 3’s 1,089.
• Speed: a 10-second video in 10.95 seconds in fal’s latency benchmark, against 277.80 seconds for Seedance 2.5, which fal says has the same reported Elo. That is the “22x faster” headline.
• Against FLUX 3: 11.5x faster, from 126.44 seconds to 10.95.

The quality figure comes from public human-preference voting, but fal calls it preliminary, and overlapping confidence intervals mean “joint top tier” rather than a clear first place. The speed numbers are fal’s own measurements on its own infrastructure.

Availability

FLUX 3 Max is available on fal’s API from 19 October for text-to-video, image-to-video and keyframes-to-video. fal says it is 75% off for the first three weeks. fal also says this is the “first iteration” and that post-training continues, so expect the model behind the name to change.

Why it matters

The quotes from both CEOs make the business point: BFL builds a foundation model, and an inference host post-trains and optimizes it into its own product. Robin Rombach of BFL says FLUX 3 was built “as a foundation for others to build on.” For video tools, where generation time decides whether iteration is pleasant or painful, an 11-second clip is a different workflow from a five-minute one.