FLUX 3 API: Access, Pricing, and What to Use Today

Black Forest Labs' own FLUX 3 model page still carries a COMING SOON label above the model name. That is the practical answer for anyone scoping an integration this week: there is no public FLUX 3 API. There is an invite-only early access program, and it currently covers video only.

What you can actually call today

Three Black Forest Labs surfaces tell a consistent story, all checked on July 27, 2026.

The launch post from July 23 describes four capabilities on separate timelines. The model page shows COMING SOON and a request form instead of a Get Started button. The pricing page has no FLUX 3 entry at all.

A single generation runs up to 20 seconds with synchronized audio. BFL's own benchmark clips were 10 seconds at 720p. Longer output comes from chaining clips rather than one long render.

The pricing page is the clearest signal for anyone building a cost model. Its calculator covers FLUX.2 flex, pro, max, klein 4B and klein 9B, plus Outpainting, Eraser and VTO. The tabs across the top read FLUX.2, FLUX Tools and FLUX.1. FLUX 3 appears nowhere except as a footer navigation link.

That means no per-second video rate, no per-megapixel image rate, and no credit conversion for FLUX 3 exists to plan against.

Inside the FLUX 3 early access application

The Request early access button on the model page opens a Tally form titled "FLUX 3 Early Access Request." It takes about a minute.

Six fields, three of them required:

• First name, last name, email (required)
• Requesting for (required): Myself or My company
• Job title / role (optional)
• What will you use FLUX 3 for? (free text)

No credit card, no company domain check, no queue position. BFL's own wording on the form is "Request free early access to FLUX 3's newest capabilities to help give feedback and share what's possible," so enrolment itself carries no charge. What it does not say is whether generation volume during early access is capped, or what happens to that usage once pricing exists.

The field that matters is the last one. BFL states it directly on the form: "Priority is given based on use case and fit, we'll follow up by email if you're enrolled." BFL does not describe the rest of its selection process, so filing early is worth doing, but the use case you write is the one input you control.

What applicants report is silence rather than a stated turnaround. @BennettWaisbren on July 27, four days after the announcement: "How do yall have access to flux 3 I've been waiting to be approved for it for like a week almost" — the duration is loose, and access requests predate the public post, but the complaint is about getting no answer. Others were generating within a day or two: @ostrisai posted clips on July 24 and @SamJWasserman reported access on July 26, using the full 20-second limit. Individual reports, not a distribution.

There is no published response time and no way to check your position, so an approval date is not something you can put on a delivery plan.

What BFL's own win rates do and don't tell you

The launch post includes preference results from 10-second, 720p text-to-video clips with audio. FLUX 3 was preferred over:

Read the bottom of that table before the top. Against Seedance 2.0 and Gemini Omni Flash the result is 52%, which is close enough to even that it should not drive a decision. BFL publishes no sample size or confidence interval with these figures, labels the whole set preliminary, and expects the numbers to move during early access.

The comparison also measures preference only. Latency, failure rate, throughput under concurrency and moderation behaviour are absent, and those are usually what decides whether a video model survives contact with a production queue. Nothing here justifies pausing delivery until the FLUX 3 API opens.

Ship this week while the FLUX 3 API is gated

Split the decision by what you need to produce.

Still images. FLUX.2 is the shipping generation of the same family, priced today. From BFL's calculator on July 27, FLUX.2 [pro] at 1024×1024 costs $0.030 per image. The published rates:

At 10,000 images a month, the gap between pro and klein 4B is roughly $300 versus $140, so test klein on the volume tier of your pipeline before assuming you need pro. FLUX.2 is callable directly from BFL with an API key from its dashboard, or through a gateway such as FLUX.2 Pro if you would rather not add a separate billing relationship for one model.

Video with sound. This is the capability FLUX 3 is being anticipated for, and the one you cannot wait on if you have delivery dates. Two models generate video with native audio today and publish what they charge. Seedance 2.0, the model BFL's own numbers put at 52%, bills by the second, and Veo 3.1 bills by the call. Gateway rates on Seedance 2.0 as of July 27 run $0.065 per second at 480p, $0.165 at 720p, $0.365 at 1080p and $0.73 at 4K, while Veo 3.1 is $1.25 per call; rates direct from each vendor differ, so check both before committing. Either way the 20-second clip FLUX 3 is headlined on costs roughly $3.30 today at 720p. That is a number you can put in a budget, which is more than FLUX 3 offers.

20-second single takes. Per-generation limits are shorter on most current video APIs, so chain clips with a consistent reference image instead. BFL describes the same approach for going beyond 20 seconds with FLUX 3, calling it "agentic chaining of individual clips into longer, multi-shot sequences."

Build the swap in now, not later

The integration shell is safe to build. The model adapter is not.

Put the concrete work first. As a general precaution with generation APIs, download outputs immediately and re-serve them from your own storage, since delivery URLs on these services are often short-lived and not meant to be handed to end users. Split retries into two policies, because a 429 should back off and retry while a moderation rejection should not. Log cost per request keyed by model ID so a swap produces a comparable number rather than a mystery.

Then keep an assumptions register instead of guessing in code. One row per unknown, with the source link and the date you checked it:

Nothing about FLUX 3 belongs in a confirmed column today.

The one thing to avoid is hardcoding a plausible-looking identifier. A family naming pattern is not a contract, and a FLUX.2 endpoint tells you nothing about the shape of the FLUX 3 one.

FAQ

Is there a FLUX 3 API right now?

No. As of July 27, 2026 there is no public FLUX 3 API, no published model ID and no endpoint documentation. Video access is granted individually through an application.

How much will the FLUX 3 API cost?

Unknown. BFL has published no rate for FLUX 3 in any unit, and the model is absent from its pricing calculator. FLUX.2 bills per megapixel and runs $0.014 to $0.07 for the first megapixel, but a video-and-audio model is unlikely to reuse a per-megapixel unit, so treat that range as unrelated rather than indicative.

How do I get FLUX 3 early access?

Apply through the form linked from the FLUX 3 model page on bfl.ai. Six fields, about a minute, no payment details. BFL says priority goes to use case and fit, and follows up by email only if you are enrolled, so no reply is not the same as a rejection notice.

Will FLUX 3 have open weights?

BFL has confirmed a planned open-weight release called FLUX 3 Dev, covering the multimodal backbone for image, video, audio and action prediction. No date, license terms or parameter count has been published.

Can I use FLUX 3 in ComfyUI?

Not yet. ComfyUI workflows need downloadable weights, and FLUX 3 Dev has not shipped. Current FLUX ComfyUI workflows run on FLUX.1 and FLUX.2 checkpoints.

Related reading

• FLUX 3 Video: what it does and when you can use it
• Seedance 2.5 API guide