FLUX 3 Max: 10-Second Video in 12 Seconds, Pricing and How to Use It
FLUX 3 Max is fal Research's new post-trained video model, built with Black Forest Labs on FLUX 3. It generates a 10-second 720p clip in about 12.45 seconds, scores 1,139 Elo in Artificial Analysis's preliminary image-to-video arena, and costs $0.0425 per second of video at its 75% launch discount through November 7, 2026. You can use it today through fal's API or playground. aiwith and Router do not host FLUX 3 Max.
Last checked: October 20, 2026. This is an access, pricing and speed guide based on fal's model page and FAQ, not a hands-on creative review. fal dates its model information to October 18, 2026 and says it is still post-training the model.
What is FLUX 3 Max?
FLUX 3 Max is not a from-scratch model. It starts from FLUX 3, Black Forest Labs' multimodal model, and fal Research continues training it for video. fal describes three layers of work:
- Post-training with Black Forest Labs to improve visual fidelity, motion coherence, prompt adherence and image-to-video quality.
- Inference optimization to cut generation time without giving up quality.
- Systems engineering on fal's own serving stack so the speed holds in production.
The result is a narrower, faster model than the base FLUX 3 release. At launch, fal lists two endpoints: text to video and image to video, both at 720p. The base FLUX 3 model is the one with draft mode, keyframes, video continuation and up-to-4K output; FLUX 3 Max focuses on the two most common ways to start a clip.
The important distinction: FLUX 3 Max is a fal post-trained variant, not the same model as FLUX 3 in BFL's dashboard. If you need a specific FLUX 3 feature such as end-frame control or video editing, check the base model's endpoints instead of assuming Max carries every option.
How fast is FLUX 3 Max?
fal's headline claim is a 10-second clip in 12.45 seconds, measured as an average for 10-second, 16:9 video at 720p/768p. That is close to real time, and it changes how you work: you can iterate on a shot in the same session instead of waiting minutes per take.
fal compares it with these models:
| Model | Image-to-video Elo | Time for a 10s video |
|---|---|---|
| FLUX 3 Max | 1,139 | 12.45 s |
| MiniMax H3 (768p) | 1,141 | 83.26 s |
| Seedance 2.5 | 1,139 | 277.80 s |
| Gemini Omni Flash 1.1 | 1,134 | 47.37 s |
| Wan 3.0 | 1,127 | 180.16 s |
| FLUX 3 | 1,089 | 126.44 s |
The speed claim is concrete: FLUX 3 Max is about 22x faster than Seedance 2.5 at the same reported Elo, and about 10x faster than FLUX 3 while gaining 50 Elo points.
Those timings come from fal's own serving setup. Your latency will depend on queue depth, region, resolution and clip length. A 10-second clip is the benchmark; shorter clips should return faster, but do not assume linear scaling.
Is the quality claim reliable?
FLUX 3 Max scores 1,139 Elo on Artificial Analysis's preliminary AA-Video-I2V v1.1 arena. fal describes it as tied for first because its confidence interval overlaps the leaders, while MiniMax H3 is listed first at 1,141. In other words, the gap between the top models is inside the noise band, and the arena is still preliminary.
That is still a strong result for a model this fast, but keep the usual caveats:
- The arena is a blind human-preference snapshot, not a task-specific benchmark.
- The comparison uses 720p/768p clips, so it says nothing about 4K behavior.
- The Elo table is vendor-published, even though the underlying arena is third-party.
For product work, test on your own prompts: faces in close-up, fast motion, text rendering, hands, product packaging and brand colors. Those are where video models usually break, and they are not fully represented by a single Elo number.
How much does FLUX 3 Max cost?
fal bills per second of output video, with no minimums and no subscription. Text-to-video and image-to-video cost the same.
| Resolution | Launch price (75% off) | List price |
|---|---|---|
| 720p | $0.0425 / second | $0.17 / second |
The 75% discount runs until November 7, 2026. A 10-second clip therefore costs about $0.43 at the launch rate, versus $1.70 at list price. The discount is a promotion, not the permanent rate, so budget from $0.17 per second if you are planning recurring production.
There is no free tier listed on the model page. fal's playground lets you try prompts, but generation is billed through your fal account.
How do you use FLUX 3 Max?
The model is available through fal's API and playground. The launch endpoints are:
blackforestlabs/flux-3-max/text-to-videoblackforestlabs/flux-3-max/image-to-video
A minimal JavaScript call using fal's client:
import { fal } from "@fal-ai/client";
const result = await fal.subscribe("blackforestlabs/flux-3-max/text-to-video", {
input: {
prompt:
"A single-file line of longhorns swims a cold jade river at dawn, only heads and wide pale horns above the water, riders on the far bank. Wide, still, enormous.",
},
logs: true,
});
console.log(result.data.video.url);
The Python client follows the same submit-and-poll pattern: send the request, receive a request ID, then poll until the clip is ready. fal's page includes both examples, and the text-to-video and image-to-video endpoints share the same billing model.
For image-to-video, pass a start image and describe the motion, camera move and lighting rather than restating the whole scene. For text-to-video, write the shot as a continuous action with a clear camera instruction; fal's own examples use one continuous shot, a named lens or camera move, and explicit "no captions, logos or watermarks" constraints.
Can you use FLUX 3 Max on aiwith or Router?
No. Neither aiwith nor Router lists FLUX 3 Max, so a subscription or API key on either service does not unlock it. If you want to make a video on aiwith today, use its existing AI video generator, which runs a different model:
- AI Video on aiwith generates 6-second clips from text or a start image at 720p by default, with 1080p on select models. It costs 60,000 beans for a silent clip or 120,000 beans with sound, and rendering takes about 2-5 minutes. This is not FLUX 3 Max and will not match its speed or motion quality.
- AI Image on aiwith is useful for the still frame you feed into an image-to-video model. It runs Nano Banana Pro, Seedream 5.0, GPT Image 2.5, Qwen Image and FLUX (the image model, not FLUX 3 Max).
If you need text for scripts, shot lists or prompt variants, Router already carries long-context models with 1M-token windows, including kimi/kimi-k3, qwen/qwen3.8-max, glm/glm-5.3 and deepseek/deepseek-flash. The current list is the authoritative place to check: router.bumo.cc/api/v1/models.
The practical split is: use Router or aiwith chat to write and refine the shot, use aiwith image generation for a start frame, and go to fal directly if you specifically need FLUX 3 Max's speed.
Should you use FLUX 3 Max?
FLUX 3 Max is worth testing if your bottleneck is iteration speed: social clips, ad variants, storyboards, concept reels, or any workflow where you want twenty takes instead of one. The 12-second generation time is the product, not a side note.
Before you build on it, keep four caveats in mind:
- It is a 720p model at launch. fal lists text-to-video and image-to-video at 720p. Do not assume 4K or the base FLUX 3 feature set carries over.
- The audio question is unresolved on this page. FLUX 3 supports native audio; fal's FLUX 3 Max page and FAQ focus on video generation and do not list audio support. Test it before you promise sound.
- The price is promotional. The $0.0425/second rate ends November 7, 2026. Plan at $0.17/second.
- The quality ranking is preliminary. Elo differences at the top are small, and the arena does not cover every production failure mode.
If you want a fast video model with an established UI and bean-based billing, start with aiwith's AI Video. If you want FLUX 3 Max specifically, start with fal's playground, run your own prompts, and compare it with FLUX 3 before committing a production pipeline.