FLUX 3 Max on fal: Joint No. 1 Image-to-Video Claim, 75% Off Until November 7

On October 19, 2026, fal announced FLUX 3 Max, a new video model from fal Research built with Black Forest Labs. fal's model page calls it "frontier video generation" optimized for speed, claims a joint No. 1 spot in a preliminary Artificial Analysis image-to-video evaluation, and prices it at a 75 percent launch discount until November 7.

The headline is a vendor claim, so this post separates three layers: what fal's own documentation states, what third-party leaderboards are reported to show, and what only testing can settle. If you are new to the model family, start with our guide to the open-weight FLUX 3 image model.

TL;DR: FLUX 3 Max at a glance

What fal and Black Forest Labs actually shipped

FLUX 3 Max sits on top of Black Forest Labs' FLUX 3 video foundation model, which fal has offered since August across eleven endpoints, as described in fal's guide to accessing FLUX 3. That guide lists standard FLUX 3 video at $0.17 per second at 720p and $0.29 at 1080p. The Max variant's regular price matches those numbers, which tells you the discount is the only thing that makes it cheaper than its parent for now.

According to the model page, FLUX 3 Max is "fal Research's frontier video generation model, post-trained in collaboration with Black Forest Labs," combining visual quality improvements with optimized inference. The text-to-video endpoint takes a prompt and optional duration, aspect ratio (auto, 21:9, 2:1, 16:9, 4:3, 1:1, 3:4, 9:16), audio toggle, seed, and a reasoning level of low or high, with high as the default. High reasoning means the model expands your prompt before generating, which fal says increases generation time.

The sample on fal's page is a 10-second, 1280 by 704 clip at 24 fps with 241 frames, and the response JSON on the page shows inference timing of about 9.5 seconds. That is one sample, not a benchmark, but it is consistent with the "about 11 seconds" claim.

Video: fal

The joint No. 1 claim, read carefully

fal's wording is precise, and worth reading in full. The page says: "In the preliminary Artificial Analysis image-to-video evaluation, FLUX 3 Max ranks joint #1, alongside the highest-ranked models." It then adds a caveat in a section titled "About the benchmarks": the result "describes the image-to-video evaluation, rather than a separate text-to-video ranking," and the speed comparison "describes the tested benchmark configuration."

So the verified statement is narrower than a headline suggests:

• The ranking is image-to-video only, not text-to-video.
• It is preliminary, and "joint" means statistically tied with other top models, not alone in first.
• The result is reported by fal, citing Artificial Analysis. We did not find the leaderboard entry independently at the time of writing, so check the Artificial Analysis video leaderboard directly.
• fal's image-to-video page adds that rankings "can change as evaluations receive more votes," so the position may move.

The speed claim: 10 seconds of video in about 11 seconds

fal says the model generates a 10-second clip in approximately 11 seconds, "more than 25x faster than Seedance 2.5 in the tested comparison." Speed claims in video are only comparable if resolution, audio and reasoning level match. fal does not spell out the comparison settings on the page, and says real time varies with request settings and operating conditions. Treat 25x as an upper-bound marketing figure for a particular configuration. The more useful reading is that a draft-like workflow, where you try several prompts quickly, becomes cheap in time as well as money.

That fits the pattern fal has been building. Its earlier FLUX 3 lineup had a draft endpoint at $0.06 per second for quick previews, and we checked a related fal pricing claim in FLUX Video Upscale 4K price cut: $0.59 to $0.28 per second.

Pricing: what the 75 percent discount really means

The model page states: "Video costs $0.0425 per second at 720p and $0.0725 per second at 1080p," and that these are promotional rates, 75 percent off, running from October 15 to November 7. Afterwards the rates are $0.17 and $0.29 per second.

These are our own multiplications of the published per-second rates, rounded to cents, and they assume one successful generation with no retries. Whether this is cheap relative to rivals is something we cannot verify, because rival prices change often. Compare against the current price of Kling and Wan before committing to volume; our coverage of Kling 4.0, Alibaba's Wan 3.0 and Grok Imagine Video pricing and Artificial Analysis rank gives reference points.

How to try FLUX 3 Max

The endpoints follow the same pattern as other fal models. From fal's FLUX 3 guide, the standard call looks like this, with the endpoint name swapped to the Max path:

import { fal } from "@fal-ai/client";

const result = await fal.subscribe("blackforestlabs/flux-3-max/text-to-video", {
  input: {
    prompt: "A continuous tracking shot follows a cyclist through a quiet coastal town at sunrise.",
    duration: 10,
    resolution: "720p",
    generate_audio: true,
    reasoning_level: "low"
  },
});
console.log(result.data.video.url);

Parameter names above come from the model page's description (duration, aspect_ratio, generate_audio, reasoning_level, seed); check the Schema tab for the exact accepted values, and for the image-to-video and keyframes endpoints. We did not run a paid generation for this post, so our verification covers the documentation, not output quality.

A sensible test plan:

1. Run the same 10 prompts at low and high reasoning and compare the prompt adherence you actually see.
2. Use one still image for image-to-video, since that is where the ranking claim applies.
3. Time each run, and note your own latency, since queueing is not in fal's benchmark.
4. Budget the discount window: until November 7, tests are 4 times cheaper than they will be afterwards.

Who should care, and who should wait

Worth testing now: teams doing rapid iteration on ads, storyboards and social clips, where seconds per render and cents per second both matter. The discount window makes this the cheapest time to benchmark.

Wait for independent results: anyone choosing a production video model based on rankings. A preliminary, joint-first image-to-video result is a reason to test, not a reason to switch. Watch for the full Artificial Analysis entry, text-to-video numbers, and hands-on reports about audio quality, motion artifacts and prompt expansion side effects.

Not for you if you need open weights: fal's page presents FLUX 3 Max as an API model. Black Forest Labs' open-weight work so far has been on the image side with FLUX 3 Klein.

Why a post-trained variant matters

FLUX 3 Max is not a new foundation model. fal describes it as built on Black Forest Labs' FLUX 3 foundation model with further post-training by fal Research. That split of labor is increasingly common: the original lab trains the base, and a serving platform with heavy inference expertise tunes it for its own hardware and latency targets. The practical consequence is that the speed numbers are tied to fal's infrastructure. If you run FLUX 3 elsewhere, do not assume the same 11-second figure.

It also explains the pricing symmetry. Because the regular Max rates equal the standard FLUX 3 rates, fal is effectively offering a faster, claimed-better model at the same list price, and using the discount to pull trial traffic. For a buyer, that lowers the risk of switching: if Max disappoints on your prompts, you can fall back to the standard FLUX 3 endpoints at the same per-second cost, using the same client code and the same output format.

Prompt expansion is the other design choice to watch. The high reasoning default rewrites your prompt before generation. That tends to help short, vague prompts and can hurt tightly controlled ones, where you want the model to follow your wording literally. Test both settings on the prompts you actually use, and keep the low setting for pipelines where reproducibility with a fixed seed matters.

What is still unknown

• The full methodology and sample size behind the Artificial Analysis evaluation.
• Text-to-video ranking. fal explicitly says the joint No. 1 does not cover it.
• The comparison settings behind the 25x speed figure.
• Whether the price after November 7 stays at $0.17 and $0.29, or the model gets a different tier.
• Long-clip consistency at 15 to 20 seconds.

We will update this post when independent leaderboard entries and developer tests appear.

Related reading

• FLUX 3 image model (Klein) open weights guide
• FLUX Video Upscale 4K price cut, checked
• Kling 4.0 launch
• Alibaba Wan 3.0 video model
• Grok Imagine Video pricing and rank

Prices, ranks and dates reflect fal's published pages on October 19, 2026 and may change.