Grok Imagine Image 2.0 vs Flux 3: The Shipped Editor vs the Multimodal Promise

This is a head-to-head where only one side is actually callable, and the honest version of that headline is the comparison itself. Grok Imagine Image 2.0, x​AI's editing-first image model, shipped in G​rok apps on August 7, 2026 — the same week it entered the public Text-to-Image Arena in second place at Elo 1,320 (preliminary), behind only GPT Image 2 at 1,380. Flux 3, Black Forest Labs' unified multimodal foundation model announced on July 23, trains image, video, audio, and robot-action prediction in one set of weights — but its image tier still has no public endpoint, no price, and no released demo as of today, August 8. One model you can call this afternoon; the other is the image release everyone has been watching since late July that nobody can run yet. That asymmetry is the whole story, and it decides what a fair comparison even looks like.

Die Anzeigetafel

On the dimensions this pairing actually turns on, here is where each side stands today. The Grok Imagine Image 2.0 figures come from the public Arena board and x​AI's own documentation; the Flux 3 image status is per Black Forest Labs' July 23 announcement and current BFL docs, and every roadmap claim on the Flux 3 side is vendor-reported and unverified until the endpoint exists.

• Availability — Grok Imagine Image 2.0 live in G​rok apps (web, iOS, Android) since August 7 vs Flux 3 image tier announced July 23, still not shipped.

• Text-to-Image Arena — Grok Imagine Image 2.0 #2 at Elo 1,320, tagged preliminary, on the public board vs Flux 3 image no ranking and no entry.

• Editing — Grok Imagine Image 2.0 ships Magic Wand, segmentation, background removal, and multi-reference input vs Flux 3 image promises editing "across styles and aspect ratios," with no released demo.

• In-image text — Grok Imagine Image 2.0 claims designer-planned typography and sharper small text vs Flux 3 image promises multilingual text rendering inside images.

• Price — Grok Imagine Image 2.0 $0.02–$0.05 per image on x​AI's API vs Flux 3 image unpriced; only the video tier ($0.06–$0.54/s) is billed today.

• API — Grok Imagine Image 2.0 developer API open, with the Image 2.0 model ID still unconfirmed vs Flux 3 image no endpoint exists.

Why this matchup can't be a normal shootout

A normal image-model comparison queues the same prompts at both endpoints and diffs the outputs. You cannot run that here, because only one endpoint exists. Grok Imagine Image 2.0 is callable through x​AI's own apps and developer API — the developer API has been live since the March generation of the base image model, and Image 2.0's dedicated API model ID is still unconfirmed, a point that matters for anyone planning around it. Flux 3 image is not callable anywhere: BFL's own docs currently show only "Video with synchronized audio, live now," with a note that "more modalities ship on the same request shape as they land." So the honest head-to-head is about what each side has shipped, what it has promised, and what you can build on today. That is a real comparison; it just isn't a benchmark one.

The editing toolset is where Image 2.0 is already ahead

Grok Imagine Image 2.0 ships editing as a first-class feature, not a follow-up. The Magic Wand edits only the region you select, leaving the rest of the frame untouched; segmentation lets you recolor or replace a precise area down to a single object's material; background removal exports a subject on a transparent canvas for reuse elsewhere; and multi-reference input accepts up to five images in one generation, so compositing several subjects does not require manual cut-and-paste. Smart Resize recomposes across nine aspect ratios, from 1:2 tall banners to 2:1 wide ones, auto-filling the frame when the dimensions change. It is trained specifically for photography, graphic design, and illustration, and it ships with prebuilt templates for product shots, headshots, e-commerce listings, game assets, and marketing posters — a positioning aimed squarely at deliverable-producing commercial workflows.

Flux 3's image tier, per BFL's launch material, promises editing "across styles and aspect ratios" and a single model that handles stills and motion together. There is no released demo, no sample gallery, and no endpoint to test. On the editing dimension this isn't close right now: one is shipped and measurable, the other is a roadmap claim.

Both are chasing the same weak spot: text inside images

Legible in-image text is the hardest unsolved problem in image generation, and both vendors aimed at it in the same fortnight. x​AI says Grok Imagine Image 2.0 plans layout and typography "the way a designer would," with sharper small text in dense graphics and fewer garbled words, logos, and labels. BFL's stated improvements for the Flux 3 image tier include "multilingual text rendering inside images" — a direct answer to the same failure mode and one of the few concrete claims in the July announcement. Neither capability has been independently benchmarked yet; both are vendor claims. The practical difference is that Image 2.0's claim is testable today on your own prompts, while Flux 3's is not testable at all.

Price, when only one side has a price

Grok Imagine Image 2.0 is pay-per-image on x​AI's API: grok-imagine-image at $0.02 per image and grok-imagine-image-quality at $0.05 per image, with a 5-requests-per-second rate limit, and edits billed for both the input image and the generated output. In the consumer apps it sits behind the $30/month SuperGrok subscription, the free tier having been removed earlier this year — which is why the API, not the app, is the price anchor for anyone building with it.

Flux 3's image tier has no price, because it has no endpoint. The only published BFL numbers are for the video tier — $0.06 per second for drafts, $0.17/s HD and $0.29/s FHD for text-to-video and image-to-video, roughly double for video-to-video — which at least signals that BFL is pricing the Flux 3 family at the premium end of the market. When the image tier lands, a per-image price well above x​AI's $0.02 floor would not be surprising, but that is an inference, not a fact; the accurate statement is that BFL has given no image pricing at all.

What Flux 3 brings that Image 2.0 can't

The case for Flux 3 was never about today's stills. BFL trained one set of weights across image, video, audio, and robot action, so the "image model" is the same model that later generates synchronized speech and effects, animates supplied keyframes into 20-second clips, and — in the longer play — predicts robot behavior. If your pipeline needs a subject's identity to survive from a still into a motion piece, or wants one request shape covering every modality, that is an architectural bet a stills-and-editing product like Grok Imagine Image 2.0 does not make.

The counterweight is that Grok Imagine Image 2.0 is a real product with a real — if preliminary — Arena rank today, and Flux 3 image is a promise that has already watched its sibling tier ship: FLUX 3 Video went generally available on August 4-5, while the image tier remains in "coming weeks." Multimodal ambition is not an endpoint.

Wer sollte welche auswählen?

If you need production image editing this quarter — background removal, region edits, multi-reference composition — Grok Imagine Image 2.0 is the only one of the two you can actually adopt, and its $0.02–$0.05 per-image price makes a systematic test of your own prompts cheap enough to just run. If you are planning a pipeline that will span stills, motion, and audio off a single model, and you can afford to wait, Flux 3 image is the more interesting bet — but it is an unshipped bet, with no committed date beyond "coming weeks" and no price.

Neither model is on OrcaRouter today: x​AI distributes Image 2.0 through its own apps and developer API, and BFL ships through its own API, and neither has landed on third-party hosts yet. When they do, the two are worth running through a single routing layer — a per-image A/B on your own prompts, automatic failover to a proven generator while either model is young, and 0% markup pass-through so the price sheet you see is the provider's list price, with any cut landing the same day. That is how you adopt a two-day-old model and an unshipped one without betting a production path on either.

Das Fazit

Grok Imagine Image 2.0 vs Flux 3 is not a close race right now, because it isn't a race — one competitor is on the track. Image 2.0 ships a capable editing toolkit at a per-image price, with an Arena rank that is real but preliminary and an editing ranking that is x​AI-cited, not independently confirmed. Flux 3 image is the more ambitious architecture, still unshipped, with no price and no demo. If you need images today, the choice makes itself. If you're betting on where image generation is going, watch Flux 3's image tier — and treat every roadmap claim as unverified until the endpoint exists.