FLUX 3 vs FLUX 3 Image: one family, and which part Sume can call

FLUX 3 is the name of a whole model family from Black Forest Labs, and FLUX 3 Image is the first part you can use as an image model. BFL's overview page (read 2026-10-03) describes one multimodal family covering image, video and audio, plus a robotics model called FLUX 3 Action 7B. If you searched for FLUX 3 and landed on image editing pages, that is why.

On Sume you can call the FLUX.2 models today, not FLUX 3. Sume's image docs list black-forest-labs/flux.2-pro and flux.2-flex among the catalog rows, and no FLUX 3 row. Check the live catalog before you plan around it.

What the family page says

Per BFL's own pages, FLUX 3 Image is the image member, with an endpoint at /v1/flux-3-image, and FLUX 3 Video sits at /v1/flux-3-video. Both are named on the overview. The image pages describe multi-reference input up to ten images, bounding-box layouts and output up to 4K. I am quoting the vendor and have not run it.

What to do this week

Do not rewrite a pipeline for a model you cannot call yet. Instead, keep your prompts model-neutral, store references and masks as files, and put the model id in a single configuration value. When a FLUX 3 row appears in GET /v1/images/models, switching is a one-line change and you can compare it on your own prompts.

Why the family matters for planning

A family launch usually arrives in stages. Image came first; the video and audio parts are named but the pages I read did not give availability or prices for them. If your roadmap needs a talking-head video or a voice, plan on Sume's own video and speech routers for now, and treat any FLUX 3 mention as a future option rather than a dependency.

A model-id check

This snippet lists every catalog row whose id contains flux and prints its reference limit. Run it when you hear a new version has launched; an empty result for flux.3 is the honest answer.

import os, requests

r = requests.get("https://api.sume.com/v1/images/models",
                 headers={"Authorization": f"Bearer {os.environ['SUME_API_KEY']}"}, timeout=30)
r.raise_for_status()
for m in r.json()["data"]:
    if "flux" in m["id"].lower():
        refs = m.get("supported_parameters", {}).get("input_references", {})
        print(m["id"], refs.get("max"))

How to compare when a row appears

When a FLUX 3 row does show up, test it on the same three prompts you use for FLUX.2 pro: one product still, one scene with text, one edit with references. Write the prompts down now, with the references, so the comparison is repeatable. Record price per image from the endpoint record rather than the vendor's list price, since Sume bills from its own endpoint pricing.

Also note that BFL's price for FLUX 3 Image is megapixel-based and a 50 percent launch discount was reported by the press through October 8; BFL's own docs pages I read did not state that end date, so treat it as a rumor until you see it on a BFL page.

Reading a vendor launch without over-claiming

Launch posts blur three things: what the model can do, what the API exposes and what a reseller carries. BFL's pages describe the model and its own endpoints. Sume's docs describe what Sume routes. A feature such as bounding-box layouts exists on the first list and is absent from the last two until a catalog row says otherwise, so a Sume user cannot send boxes today and should not read the launch as a promise.

Open weights and timing

BFL says open weights are expected but gives no date in the pages I read, so do not plan a self-hosted deployment around it. For how FLUX.2 behaves at wide ratios, see FLUX.2 21:9 versus FLUX 3, and for which Sume model suits text in images, the text-in-images guide. The catalog is documented in the Image API docs.