FLUX 3 Max
Release date unpublished
Speed-oriented post-train of FLUX 3, built by fal Research with Black Forest Labs on the FLUX 3 base. It generates text- and image-to-video with native audio, a 10-second clip in about 12 seconds.
- Price per second
- $0.17
- Max clip length
- 20 s
- Max resolution
- 1080p
- Native audio
- Yes
- AA arena Elo
- 1155
- LMArena Elo
- —
- Capabilities
- text to video, image to video
- Licence
- Proprietary
Strengths & weaknesses
Strengths
- No. 6 on Artificial Analysis' text-to-video leaderboard with audio (Elo 1155)
- 10-second 720p clip in about 12 seconds (fal-reported, down from about 126 s for FLUX 3)
- 5 to 20 s clips with audio on by default
Weaknesses
- 1080p maximum (FLUX 3 goes to 4K)
- No continuation/extension endpoint, unlike FLUX 3
Separate model from flux-3-video: fal calls it 'a fal Research model built with Black Forest Labs on the FLUX 3 base', and BFL's FLUX 3 Video page carries an 'Introducing FLUX 3 Max' section ('10-second video in 12 seconds', '22x faster than Seedance 2.5 at the same quality'). Release day not published on fal or BFL pages; AA lists 'Oct 2026', so releaseDate is null. No BFL-native endpoint found (BFL docs list only /v1/flux-3-video); ids are fal's (a keyframes-to-video endpoint also exists on fal but its id was seen only in a URL and is not recorded). fal text-to-video API: duration 5-20 s (default 5), resolution enum 720p/1080p (default 720p), generate_audio default true, reasoning_level low/high. Speed figures conflict across fal pages (12.45 s / ~22x on the post vs ~11 s / '>25x' on model pages). fal's claim of tied #1 on AA's preliminary image-to-video arena (1139 vs FLUX 3's 1089) is vendor-reported. No weights mentioned anywhere, so openWeights false. Not on LMArena's video board. Notable by clause 1: AA-Video-T2V v2.1 #6 as 'FLUX 3 Max (720p)' (2026-10-27).
Links
API model strings
- fal
blackforestlabs/flux-3-max/text-to-video - fal
blackforestlabs/flux-3-max/image-to-video