news

Adobe expands Firefly with agents and partner models at MAX 2026, and names no new Firefly model version

The November 10 keynote release adds Creative Agent, Firefly Graph and Google's unnamed "latest" image and video models, with no quality or speed figure attached to any of it.

By Priya Shenoy · 11 November 2026

Adobe's largest Firefly announcement of the year, delivered on day one of MAX 2026 in Miami Beach, is about agents, reusable workflows and other companies' models. It names no new version of Adobe's own Firefly image or video model, and neither of the two newsroom posts puts a benchmark score or a generation time next to anything it announces.

What Adobe put into Firefly on November 10

Adobe described the update as expanding Firefly into an AI creative studio with new agentic capabilities, what it calls "top industry models" and new tools across the video, design and image editors. The Adobe Creative Agent is being extended across Firefly, with new capabilities in the Firefly AI Assistant, which remains in beta. Firefly Graph, also in beta, lets creators build repeatable workflows that chain AI models together with Adobe's own product tools, and Firefly Builder is being previewed in private beta as a way to turn a written description into an interactive prototype.

On the Creative Cloud side, Adobe said the Creative Agent is spreading into apps including Premiere, Illustrator, After Effects and InDesign. TechCrunch reported two further pieces from the same event: a Firefly video tool that assembles a video out of clips, images and sound effects, and a local Model Context Protocol server that exposes Creative Cloud features to outside assistants such as ChatGPT, Claude or Gemini.

The partner models arrive by company name, not by version number

Adobe's release states that customers will continue to get Google's latest image and video models in Firefly, Photoshop, Adobe Express and Premiere, and that Gemini models will help power AI features in Acrobat. The word doing the work there is "latest," which is not a version number and not a commitment to a specific model. Google's video lineup has already moved once this autumn: the three Veo 3.1 preview endpoints were switched off in October, with a Gemini Omni Flash build named as the replacement. A Firefly or Premiere user reading "Google's latest" today cannot tell from Adobe's text which of those they will be billed for tomorrow. On stage, a Photoshop segment used generations from Google's Nano Banana 2, according to live coverage of the keynote.

For the models Adobe does not build itself, the per-second economics are set elsewhere, and they vary by the platform reselling them. Our radar tracks what one of those Google models costs across the platforms that carry it.

No measured claim accompanies the launch

Across the two newsroom posts and the Creative Cloud blog entry, there is no output quality comparison, no latency figure and no stated resolution or duration ceiling for the new Firefly video tooling. That gap matters more than usual for this particular release, because the pitch is that an agent chooses and chains models on the user's behalf. When the routing is automated and the model list is described by partner rather than by version, the two things a buyer most needs to compare, what the output scores and what each generation costs, are the two things the announcement leaves out. Adobe has form here: its Firefly Foundry deal with Disney was also announced without published training details.

MAX runs through November 12, with the Sneaks session on November 11, so some of the unnamed pieces may acquire names before the week is out.

Sources: Adobe Newsroom, Adobe Newsroom, Adobe Blog, TechCrunch, Digital Camera World.

Priya Shenoy: Tracks what AI video actually costs across the platforms that resell access to the same handful of models. Treats a pricing page as a claim, not a fact, until someone checks it.