A price war is brewing between the two biggest AI labs, and small business owners stand to win.

Main

Here's something that should make you feel better about every AI bill you've paid this year: you've been subsidizing the labs.

A research firm called SemiAnalysis ran the numbers and found that Anthropic's $200/month Claude Max plan can cost Anthropic up to $8,000 in compute per heavy user. OpenAI's ChatGPT Pro plan costs them as much as $14,000 per month for the same kind of user. They've been eating the difference hoping you'd stick around.

That arrangement is ending. The Wall Street Journal reports that OpenAI is considering significant price cuts on API access, anticipating Anthropic will do the same. This is a full-on price war, and it hasn't started yet.

What that means for you: if you're building anything on top of these APIs, or you're locked into a long-term contract, don't sign anything annual right now. Costs are likely to drop, and drop meaningfully. Wait, or negotiate a short-term deal first.

---

Now, here's a real business showing what AI-generated content actually looks like when it works.

Local news company 6AM City cut 30 of its markets, kept 19 with human editors, and then launched 400 more newsletters run entirely by AI. Not partially. Entirely.

Co-founder Ryan Heafy told A Media Operator that sending each of those newsletters costs less than a dollar a day. A single performance-based ad click makes each one profitable. The company hit profitability in Q1 2026 with a 10 to 20 percent margin.

That's the number worth staring at. Not the technology, not the headcount reduction. A dollar a day per newsletter, and one click to break even.

If you're running email newsletters, local content, or any content marketing program, that math applies to your business too. The floor on content costs just fell through.

---

One more thing worth knowing this week. OpenAI's latest ChatGPT Business update added three things at once. First, you can now create and edit PowerPoint presentations directly inside ChatGPT, which matters if you're constantly building client decks. Second, workspace agents can now be triggered from your other software via API. Think of that as a remote control: your helpdesk, your scheduling tool, or your CRM can now kick off a ChatGPT agent automatically, without anyone touching a keyboard. Third, and most practically, the free period for those workspace agents runs until July 6, 2026. After that, credit-based pricing kicks in.

That's a real deadline. If you've been curious about automating lead qualification, meeting follow-ups, or vendor routing, this is the window to experiment at no cost.

New Tools

Lovable (May 2026): AI app builder now runs parallel subagents for faster builds, adds reusable "Skills" playbooks, automatic security scans before every publish, branded workspace URLs, and a Salesforce connector. For small teams building internal tools or client dashboards. Business/Enterprise plan required for branded URLs and Salesforce connector.

DigitalOcean Inference Router: Routes each API call to the cheapest model that can handle the task, so simple jobs like classification and summarization stop burning frontier-model budget. No code changes required. For anyone running AI apps at scale. Pay-per-token; no separate fee for the router.

Claude Fable 5: Anthropic's strongest publicly available model right now, with top scores in coding, long-horizon reasoning, and vision tasks. Available to paid-plan users through June 22. Included in existing Anthropic paid plans.

Google Business Profile Recurring Posts: Schedule repeating posts directly inside your Google Business Profile dashboard, no third-party tool needed. For any local business that wants a consistent posting cadence. Free.

Google Local Service Ads, Website Content Enhancement: LSAs now pull from your website to automatically enrich ad placements, going beyond static profile data. For local service businesses (plumbers, lawyers, cleaners, etc.) who want better performance without extra work. Standard pay-per-lead pricing.

More Headlines

Anthropic apologizes for secretly downgrading Claude Fable 5 responses: Researchers found that Fable 5 was silently routing "high-risk" prompts to a weaker model without telling users, Anthropic apologized and promised visible fallback indicators going forward. If Fable 5 seems weaker than expected on technical prompts, check those new transparency signals.

Only 1 in 5 U.S. companies uses AI in any business function, per Census Bureau data: A Stanford analysis found about 20% of American businesses use AI at all, which means if you're reading this newsletter, you're already in the early-mover minority.

Hiten Shah: the real AI value for owners is in the correction loop, not the first output: Worth a quick read if you've been frustrated that AI "gets it wrong" on the first try.

Apple unveils smarter Siri powered by Google Gemini at Tim Cook's final WWDC: Big news for anyone building or buying consumer-facing Apple products.

Google adds back-button hijacking to its spam policy: Site owners and plugin users who use this pattern are now at risk of a search visibility penalty.

Google Maps testing in-app video editing for local business profiles on Android: Early-stage feature, but worth watching if local video content is part of your marketing.

Claude for small business, a practical YouTube review: AI educator Alex Northstar walks through whether Claude justifies its cost versus alternatives, from an owner's perspective.

SemiAnalysis cost breakdown: Claude Max at $200/month costs Anthropic up to $8,000; ChatGPT Pro costs OpenAI up to $14,000: The numbers behind the coming price war, for anyone who wants the full picture.

DiffusionGemma: Google releases an open Apache 2.0 text model with 4x faster inference: Generates text in parallel rather than word-by-word, which matters if inference speed or cost is a constraint for your application.

-Scott

*SmartOwner is published (almost) daily by the team at DigitalTreehouse. Want AI consulting or automations for your