fal.ai debuts FLUX 3 Max video model with 22× speed claim over Seedance 2.5

On October 20, fal.ai launched FLUX 3 Max, a video model developed in collaboration with German AI lab Black Forest Labs. The company claims the model generates a 10-second, 16:9 clip in an average of 12.45 seconds, compared with 277.80 seconds for Seedance 2.5 in its benchmark—roughly 22 times faster. fal.ai says additional training focused on closer instruction adherence, better-looking output and reduced generation time.

According to the Artificial Analysis image-to-video leaderboard checked on October 19, FLUX 3 Max scored 1,138 Elo, tying with Seedance 2.5 and trailing MiniMax H3 by two points. Confidence intervals for the top models overlap, and the evaluation is based on preference tests for image-to-video with audio, not text-to-video quality. As a result, FLUX 3 Max appears competitive with leaders, but a definitive quality advantage cannot be established from these results.

FLUX 3 Max supports text-to-video and image-to-video generation, with output lengths from 5 to 20 seconds, 720p resolution (upscalable to 1080p), optional audio and multiple aspect ratios. A separate keyframes endpoint accepts up to 10 input images. fal.ai is offering a 75% promotional discount through November 7, 2026; during the promo, 720p video costs $0.0425 per output second, or about $0.425 for 10 seconds, versus a regular rate of $0.17 per second.

fal.ai says API-generated content can be used commercially, but the rights do not include downloading, modifying or running the model independently. Faster generation could reduce waiting, but may also encourage more attempts; the real value for teams depends on whether they can reliably reach an acceptable result faster at a predictable cost.