Every business owner asks me the same question before automating anything: what does this cost per month. Part of that answer moved on July 30, 2026 — and the larger part didn't move at all. Telling the two apart is the difference between a budget that holds and a contract that surprises you in month three.
What happened
OpenAI's official API changelog, entry dated July 30, 2026, states it plainly: "Starting July 30, GPT-5.6 Luna costs 80% less, while GPT-5.6 Terra costs 20% less." The cheapest model in the line got 80% cheaper; the mid-tier one, 20%.
This isn't a launch promotion. It's the continuation of a pattern that has been running for years: the price per token — the unit you pay every time a model reads or writes something — keeps falling, and falling fast, while capability climbs. The live per-model table lives on OpenAI's own API pricing page, and it's worth checking the date on it before you put any figure into a spreadsheet.
What did not move in the same window: the per-seat subscription. ChatGPT Business is still listed at $20 per user/month on the annual plan and $25 per user/month billed monthly, with a 2-user minimum. Token price and seat price are two different bills, and only one of them dropped.
What changes for you
If you run a small operation — inbound calls, quotes, follow-up, content — the token bill is almost never the biggest line in your automation budget. The lines usually rank like this:
- Seats and licenses. Per-person subscriptions, CRM, scheduling tool, telephony. Charged per head, grows with the team, and did not drop in July.
- Plumbing. Connecting the tools you already run, handling exceptions, fixing what breaks. You either pay an agency or you pay in your own hours.
- The model. The tokens themselves.
Which gives you the honest read on the news: an 80% cut on a line that is 10% of your bill moves the total by 8%, not by 80%. If a vendor tells you your bill will halve because models got cheaper, they're selling, not calculating.
Where the cut genuinely matters is in what now fits. High-volume work that never penciled out before — classifying every inbound email, drafting a first reply for every lead, summarizing every call — starts to pencil out. Not because AI got better this week, but because the floor on cost dropped.
In practice
Four concrete moves, in the order they're worth doing.
1. Split volume work from money work. Not every task needs the expensive model. Triage, classification, first drafts, summaries — volume — run on the cheap model. A proposal going to a client, copy that carries your name, a decision that touches money — money work — runs on the good one. That's the routing policy we run inside the Marcus Aurelius System, and it's what turns an 80% price cut into actual dollars at the end of the month.
2. Measure cost per delivered thing, not per token. Tokens are an input; what matters is what the finished item costs. One data point of our own, stated as an aggregate and labelled as ours: in the Marcus Aurelius System content engine, the cover image for each published piece costs $0.08, which is fal.ai's published per-image price for nano-banana-2. Posting to Facebook, Instagram and LinkedIn carries no platform cost. So the marginal cost of one published piece, in our case, lives in cents — which is exactly why the cadence is sustainable.
3. Run the example on your own numbers. An explicitly hypothetical exercise with stated assumptions: take an operation with 2 seats at $20/month ($40), $60 of supporting tools, and $20 of model usage — $120 total. An 80% cut on the model line takes that $20 down to $4. New total: $104. A 13% drop overall. Those figures are mine, invented to show the proportion, not a projection of your situation — swap in your real numbers before deciding anything.
4. Revisit pricing every quarter. Model prices moved twice this year alone. An automation contract that locks your price for 24 months locks you to an input cost the market has already left behind. Ask for a review clause.
One note that applies to us too: nobody can promise what automation does to your revenue, because nobody outside your business knows your ticket size, your volume, or your margin. Results vary with the business, the sector and the execution. What can be promised is a transparent calculation — and your ability to redo that calculation without us.
Sources
- OpenAI — API Changelog (entry dated July 30, 2026)
- OpenAI — API Pricing (per-model table)
- OpenAI — Business Pricing ($20/user/month annual, $25 monthly)
- fal.ai — nano-banana-2, price per image
One honest note: there is no universal multiplier. The outcome depends on your industry, your ticket, and how much of your volume is urgent versus price-shopping. Anyone promising a fixed return is selling, not calculating.