AI news & training for business owners & operators. One email. One clear next step.

THIS WEEK IN AI

This week, none of the AI news was about making AI more powerful. It was about who gets access, who decides, and at what price. The tier you pick, and what your team actually does on it, is now where the real money is made or lost.

What happened:

OpenAI previewed GPT-5.6 on June 26th. The new model ships in three named tiers: Sol at the top ($5 per million input tokens, $30 per million output), Terra in the middle ($2.50 / $15, with about half the cost of GPT-5.5 at similar quality), and Luna at the bottom ($1 / $6). The launch is a limited preview to trusted partners only. OpenAI says the US government asked them to gate access while a cyber Executive Order framework gets worked out, and they openly say they don't want that to become the long-term default.

Why it matters to your business:

The middle tier just got a lot cheaper. If your team is on the top tier of GPT-5.5 and using it for meeting notes, email drafts, and simple summaries, you are likely overspending by 50% or more on that work. The bigger shift is the naming. Sol, Terra, Luna gives you a real ladder for matching tasks to cost. The buying decision is moving from "which AI" to "which tier of which AI." Get familiar with that question now.

Source: openai.com/index/previewing-gpt-5-6-sol

What happened:

On June 22, OpenAI expanded Daybreak, its cybersecurity program. The full GPT-5.5-Cyber model is now available to trusted defenders. A new Codex Security plugin finds bugs and writes patches inside the same tool engineers use to code. And 25-plus security firms (Cisco, CrowdStrike, Palo Alto Networks, IBM, Check Point, Accenture, and others) signed on to put GPT-5.5-Cyber inside their own products. IBM and OpenAI announced a separate application security service the same day.

Why it matters to your business:

If you pay for any of those security platforms, the AI capability is coming to you through your existing vendor over the next few months. You don't have to integrate anything. The skill that pays off now is knowing what to ask your security partner: how are they using the model, where does the human review sit, and what shows up in the reports they send you. The AI is already in your security stack. You just need to know how to read it.

Source: openai.com/index/daybreak-securing-the-world

What happened:

At Google I/O on May 19, Sundar Pichai told the audience Gemini 3.5 Pro would ship "next month," meaning June. June ends in four days. Business Insider reported on June 25 that the launch slipped to July. Google has not officially confirmed the delay. The model sits in a limited enterprise preview, and Pichai's original "next month" promise reportedly drew audible groans from the I/O audience.

Why it matters to your business:

Vendor roadmap dates are not contracts. If your team is building a workflow around a model that "ships next month," budget two months. Better, build on what is available today and treat new model launches as upgrades, not unlocks. The teams that win are the ones whose workflows still function the week the roadmap slips.

Source: Business Insider via Digit

For two years, the AI question every business owner asked was "which AI should we use?" That was the wrong question. It just took a while for the pricing pages to catch up.

This week, OpenAI made the real question visible. GPT-5.6 doesn't come as one model. It comes as three: Sol, Terra, Luna. Each one is a different tier of intelligence at a different price. Same generation. Different jobs. The vendor you might already pay for AI is now asking you to pick a tier for each workflow.

Daybreak made it structural on the other side. OpenAI is no longer selling cyber AI directly to most companies. They are selling it through Cisco. Through CrowdStrike. Through IBM. Through 25 partners with their own access tiers, governance models, and audit trails. The AI is in your security stack already. You just don't see the brand on it.

And then Google missed its own deadline. The model the industry was waiting for is sitting in a small enterprise preview while Google decides who gets the keys.

Three stories. One pattern. The era of "we ship the most powerful model" is being replaced by "we ship the right tier for the right job at the right level of access." That sounds like marketing copy until you put it next to a real workflow.

Here is what it looks like in practice. A pattern I see in small offices: the whole team is on the top tier of one AI tool. Twelve seats. Maybe fifteen. The engineers use about 40% of the capability. Everyone else is using it to summarize meeting notes and draft emails. That work could have happened on the cheapest tier of the cheapest model and nobody would have noticed a difference. The team is paying for capability that is not getting touched.

The reason that happens is not that owners are bad at math. It is that there was no clean tier to drop down to. Now there is. And by the end of summer, every major lab will have one.

So the operator move is this. Stop asking "which AI should we use?" Start asking "which tier of which AI matches which workflow?" That is a different question. It pushes the decision down to the work, not up to the vendor.

I'll be honest. I am not fully convinced any of the labs have the tier story figured out yet. The names will keep changing. The pricing will move. Some of the cheaper tiers will turn out to be a bad fit for serious work, and some of the expensive ones will be overkill for everyone. But the direction is set. The buying decision is moving from the model to the workflow.

If you are signing an AI contract this quarter, don't negotiate the seat price. Negotiate the tier flexibility. Make sure your contract lets you move people up and down as you figure out where the work actually sits. That is where the savings will come from over the next 18 months. Not from picking the cheapest vendor. From picking the cheapest tier that still gets the job done.

One question for your next leadership meeting:

"For every AI tool we pay for, who is on the top tier and what work is actually getting done there?"

If your team can't answer in one sentence per seat, you are buying the brand, not the workflow. The fix is small. Pull your last 60 days of AI invoices. For each seat, ask the workflow owner what that person is actually using AI for. Write it down in one sentence. Mark each sentence as "top tier work" or "everyday work." Then move the everyday work to a cheaper tier or a cheaper tool.

Most of the savings will sit on the everyday work side. Meeting notes. Email drafts. Simple summaries. That work runs fine on the cheaper tiers that just got cheaper this week.

Forward this to your operations lead. Ask them to walk into the next meeting with the answer.

If your AI invoices are climbing and you are not sure where the tier decision should sit, that is the work we cover on a paid AI Discovery Call. 30 minutes. You leave with a clear read on where the spend is going and where it should sit.

Book your discovery call here: DISCOVERY CALL LINK

Thanks for reading,

Andrew Keener

Operations & AI Strategist

Keen Alliance Consulting