July 2026 delivered the densest wave of AI model releases in a single month so far — and unusually, the headline wasn’t capability. It was price.
Images at roughly three cents. Video from five cents a second. Text models dropping 80% in a single announcement. For a small marketing team, that changes something concrete: creative volume is no longer a budget decision.
Here’s what actually shipped, what it costs, and the two constraints that got tighter rather than looser.
Text: cheaper and better in the same month
OpenAI released the GPT-5.6 family publicly on July 9 — Sol as the flagship, Terra as the balanced option, Luna as the low-cost tier — following a limited preview on June 26 that was restricted at the request of the U.S. government. Then on July 30 it cut prices: Luna by 80%, Terra by 20%.
Anthropic released Claude Opus 5 on July 24, described by Gizmodo as intelligence approaching its top-tier Fable 5 model at half the price, with a 1M-token context window and up to 128K output. It became the default on Claude Max.
The context window matters more than it sounds for marketing work specifically. A million tokens means an entire research corpus — every competitor’s site, a year of your own content, a full set of transcripts — analysed in one pass rather than chunked, summarised and stitched back together with errors accumulating at each seam.
OpenAI also announced GPT-Live for voice, ChatGPT Work, OpenAI Presence for enterprise agents (July 22), and Health in ChatGPT (July 23).
Images: three cents and four seconds
Google released Nano Banana 2 Lite (Gemini 3.1 Flash-Lite Image) on July 1: generation in roughly 4 seconds at about $0.034 per 1K image. Alongside it came Gemini Omni Flash, handling video and conversational editing at $0.10 per second. Both embed SynthID watermarking. On July 14, Google added image generation directly inside AI Overviews.
Run the arithmetic, because it’s the whole point:
Twenty image variants for an A/B test: about $0.68.
Not twenty concepts — twenty executions of a concept. Different backgrounds, different crops, different colour treatments, different text placements. The cost of finding out which one works is now indistinguishable from zero.
The old creative process was shaped entirely by the fact that each execution was expensive. You picked two options because two was what the budget allowed, and you defended the choice with judgment because you couldn’t afford to test it. That constraint is gone, and most teams haven’t restructured around its absence.
Video: three viable options, one to abandon
As of July 2026, the practical field:
Veo 3.1 (Google) — the safe institutional choice, from around $0.05 per second. Kling 3.0 — the best price-to-quality ratio at roughly $0.10 per second. Seedance 2.x — the strongest option for longer image-to-video sequences.
And one deadline: OpenAI has confirmed Sora is shutting down. The app closed on April 26, 2026; the API follows on September 24, 2026.
If any part of your production pipeline depends on Sora, you have roughly seven weeks. Don’t start anything new on it.
For social content on a small budget, Kling and Seedance give the better economics. Veo is what you pick when a client or procurement process needs a name they recognise.
The constraint that got tighter
Everything above got cheaper. One thing got more expensive, and it took effect today.
The EU AI Act’s transparency obligations apply from August 2, 2026. The European Commission adopted its Article 50 guidance on July 20. The requirements: disclose AI interaction (a chatbot must identify itself as a machine), disclose AI-generated content, disclose deepfakes, and disclose AI-written text on matters of public interest. Penalties run to €15 million or 3% of global turnover.
The detail that catches people out: the rule applies where the advertising is shown, not where the company is registered. A Dubai agency running a campaign to an audience in Germany is inside scope.
Where the line sits: ordinary photo retouching doesn’t require labelling. Creating something realistic that never existed does.
Separately, the Digital Omnibus (Regulation (EU) 2026/1744) entered into force on July 27 and postponed the high-risk obligations to 2027–2028 — but the August 2 transparency deadline was not moved.
Note that SynthID watermarking is embedded in Google’s image and video output by default. That’s a compliance asset, not a nuisance.
The other constraint: nobody wants to see it
Cheap generation solves supply. It does nothing about reception.
Harris Poll research presented at Cannes Lions found 63% of consumers less likely to buy from a brand using AI-generated advertising, and 73% less likely to trust an ad they suspect was AI-made. Klaviyo/Datalily research across 8,000 consumers in eight countries found only 7% say visible AI content increases brand trust, while 31% say it decreases it.
So the winning configuration isn’t «more AI content.» It’s more AI behind the content: variant generation, research at volume, first drafts, production systems, background and asset work — with a human making the final call on everything the audience actually sees.
The teams that lose here will be the ones that treated a 95% cost reduction as permission to publish 20× more.
What to do this month
Restructure creative testing around volume. If you’re still shipping two versions of an ad, you’re running a process designed for a cost structure that no longer exists. Ship twelve. Let the data pick.
Get off Sora before September 24. Move to Kling or Seedance and rebuild any dependent workflow now, while it’s a planned migration rather than an outage.
Audit your EU exposure today. If realistic AI-generated content reaches an audience in the EU — including diaspora audiences you may not think of as «European» — labelling is required as of today.
Keep the visible layer human. Voice, point of view, faces, final judgment. Use the savings to buy more thinking time, not more output.
The bottleneck moved
For most of the last two years, the honest constraint on marketing output was production capacity: you couldn’t make enough, fast enough, cheaply enough.
That’s over. What’s left is knowing what to make, and having the taste to recognise which of the twelve versions is actually good.
Those didn’t get cheaper. If anything, they just became the only thing you’re really paying for.
About the author
Alina Palii — brand strategist, founder of ALPA Marketing.
She works with founders and leadership teams on the decisions that come before the marketing: what the brand stands for, who it is genuinely for, what it declines to be, and how that translates into everything the market eventually sees. Strategy first — the content, the channels and the campaigns follow from it.
10+ years in marketing and a master’s degree in the field. She has built brands from zero for AI startups and national companies, and shaped the positioning of personal brands whose audiences buy on trust rather than on price.
Ukrainian by origin, living between Dubai, Paris and Ukraine, and working across the UAE and European markets — a vantage point that matters when a brand has to hold its meaning across cultures rather than be rebuilt in each new one.
She works across categories rather than inside one. Positioning logic travels between industries even when the audience doesn’t, and the pattern recognition that comes from moving between them is often what a category-blind competitor is missing.
Alina takes on a limited number of strategy engagements at a time.
