news

FLUX 3 Max ships with three different timings for the same 10-second clip

fal and Black Forest Labs put the same generation at 12, 12.45 and 10.95 seconds, and the 22x speed claim is pinned to a 4,868-vote arena Elo the release itself calls preliminary.

By Priya Shenoy · 20 October 2026

FLUX 3 Max was announced on October 19, 2026 with one number doing nearly all the work: a 10-second video generated in roughly 12 seconds, which the launch frames as 22 times faster than Seedance 2.5 at the same quality. Across the announcement materials that generation is timed three separate ways, and the quality half of the comparison rests on an arena rating the release describes as preliminary.

fal post-trained Black Forest Labs' model, and both companies announced it

The model comes from fal rather than from Black Forest Labs directly. fal's release describes FLUX 3 Max as post-trained by fal Research in collaboration with Black Forest Labs, built on the FLUX 3 foundation model through post-training, reinforcement learning and inference optimization. BFL CEO Robin Rombach is quoted saying FLUX 3 was built "as a foundation for others to build on," with fal customizing it for developers on fal's platform. fal CEO Burkay Gur frames the result as optimizing models and the systems running them together.

Black Forest Labs announced the same model on its own account, with the claims compressed into four lines: fastest in the west, a 10-second video in 12 seconds, tied for #1 in image-to-video, and 22x faster than Seedance 2.5 at the same quality.

Twelve seconds, 12.45 seconds, or 10.95 seconds

The headline figure is 12 seconds. Further down the same BFL thread, the number becomes 12.45 seconds, now specified as 22x faster than Seedance 2.5 at the same Elo rating rather than the same quality. fal's press release gives a third figure for what reads as the same test: in fal's latency benchmarks, a 10-second video in 10.95 seconds, over 22x faster than Seedance 2.5. None of the three is published with the hardware, the resolution, the prompt, or whether the run was text-to-video or image-to-video, which are the variables that decide whether any of them can be reproduced.

The parity claim comes from 4,868 votes

The "same quality" half of the comparison is a single external data point. fal reports that in preliminary independent human preference evaluations on Artificial Analysis Video Arena, FLUX 3 Max scored 1,139 Elo across 4,868 votes, which it calls placing the model among the highest-quality image-to-video models evaluated. BFL's version of that result is firmer, stating a tie for first place. Neither publishes a confidence interval, and at this vote count a tie claim and a lead claim are usually the same measurement read two ways. fal separately claims 11.5x faster inference than the original FLUX 3.

In our own benchmark, FLUX 3 and Seedance 2.5 are the base models on either side of that comparison, and the cost gap is the other half of the pitch.

A discount is stated, a price is not

BFL's thread says FLUX 3 Max is 75% off for two weeks, calling it the highest-quality, fastest and cheapest model for overall video generation. The announcement does not state what the price is before or after that discount, so the cheapest claim is the one number in the launch that cannot be checked against the launch itself. Access is through fal, with separate text-to-video, image-to-video and keyframe endpoints.

Sources: fal (PR Newswire), Black Forest Labs.

Priya Shenoy: Tracks what AI video actually costs across the platforms that resell access to the same handful of models. Treats a pricing page as a claim, not a fact, until someone checks it.