ChatGPT Images 2.5 Just Cut Latency in Half. The Brief Did Not.

OpenAI just shipped ChatGPT Images 2.5 and said generation latency dropped by up to 50 percent. Better reference fidelity. Tighter edits. Two new API models sitting next to the product.

Same 48 hours, Google started converting Search campaigns that used auto-created assets or campaign-level broad match into AI Max. Legacy setups are being switched off.

Cool week for product pages. Not the brief.

The brief is this: production got faster than the meeting that commissions it. And most marketing teams are still measuring the old click.

What actually shipped

LAYER 1 — Images stopped being the bottleneck. A sketch, a product shot, and a line of copy now get you a usable still before the client finishes “can we also try…” That is production speed, not a demo.

LAYER 2 — Paid search is becoming an agent surface. Google is not asking if you want AI Max. It is migrating the inventory. If your account still lives on last year’s settings, the platform is rewriting the campaign while you argue about headlines.

LAYER 3 — The labs are still dumping models. GPT-6 Astra. Claude Fable 5.1 and Mythos 5.1. Gemini 3.8 Flash. Meta Muse Spark. Nvidia agreed to buy Hugging Face for $12.9 billion. Model fatigue is real. The work did not wait.

What this means for marketing

BCG’s 2026 CMO survey is the quiet punchline. 96 percent of CMOs say AI is transforming marketing end to end. About a third have actually rebuilt the work. Just 8 percent run campaigns where multiple agents operate without a human assembling every step.

That gap is the agency job this month.

Generative engine optimization is no longer a side quest. Buyers shortlist inside ChatGPT, Gemini, and Perplexity before they hit your homepage. If the answer engine cannot cite you cleanly, you do not make the meeting. GEO and AEO now sit next to SEO the way paid sat next to organic ten years ago.

Do not rebuild the stack again. Do this:

• Take one live client brief. Generate 20 image variations in Images 2.5 or Firefly before the first internal review. Ship 3. Kill the rest.

• Ask ChatGPT, Gemini, and Perplexity the exact buying questions your client should win. Screenshot who gets named. That is the GEO audit.

• If you run Google Search, check which campaigns already flipped to AI Max. Write the human QA layer now, not after the spend moves.

Do this next

Pick one account. Finish one production loop this week. Image set. Answer-engine audit. Campaign check. Then send the client the before and after, not another model recap.

The labs will keep declaring eras. Agencies that ship the work will own the next one.