Most small business owners carry a quiet worry about this technology that rarely gets said out loud: that it will turn into one more subscription they pay for and barely use. So when the price of the underlying models drops by 80 percent in a single move, it is worth asking what actually changes on the ground, and whether cheaper AI tools help a five-person company or just make a big number look smaller.
The move was real and recent. On July 30, 2026, OpenAI cut the price of its GPT-5.6 Luna model by 80 percent and its Terra model by 20 percent, as reported by CNBC and VentureBeat. Luna now runs at $0.20 per million input tokens and $1.20 per million output tokens. VentureBeat noted the cut arrived just three weeks after the GPT-5.6 family launched on July 9, an unusually fast repricing for a company at the front of the field.
Why cheaper AI tools are suddenly everywhere
OpenAI is not moving alone. Google introduced Gemini 3.7 Flash for coding, document work and business automation, with introductory pricing of $0.75 per million input tokens through the end of 2026. eWeek reported that Luna, at its new rate, undercuts Google’s Gemini 3.6 Flash and Anthropic’s Claude Sonnet 5 by a wide margin. CNBC put the reason plainly: the cuts came as companies grow more sensitive to cost.
The knock-on effect is the part that matters for a small business. When tokens get this cheap, the math behind everyday automation changes. The analysis site InfiniteUp described it in blunt terms: a small firm’s entire monthly usage can now cost less than one staff lunch. That is roughly what the numbers say today, not a promise about next year.
Cheap is not the same as useful
Here is the other half, and it is the half the pricing headlines skip. A cheaper tool is only worth anything when there is a real job for it to do. Falling prices remove one excuse for not starting, but they do not tell an owner which task to point the tool at, and they do not do the work of integrating it into the day.
There is also a trap worth naming. CloudZero’s 2026 pricing analysis argues that the cheapest model that almost works is often the most expensive choice. A model one tier too small produces failures a business pays for later, in support tickets, retries and lost trust. The more useful question is not what the cheapest option is, but what the cheapest option that does this particular job well is. Those are not the same sentence.
This is the same reason so many efforts stall. Plenty of small businesses have watched a promising tool never make it out of the pilot stage, not because the model was too expensive, but because no one owned the job it was meant to do.
What a small business can do this week
None of this needs a strategy deck. The first step is small and concrete. Pick one task that repeats, carries low risk, and eats time: drafting replies to common customer questions, summarizing a week of invoices, cleaning up a messy spreadsheet. Run it on one of these cheaper models for a week and keep a simple note of what it got right and where it needed a person.
Lower prices help most in exactly these high-frequency, low-drama corners. Work like routine bookkeeping is where the falling cost quietly compounds, because the same task runs hundreds of times a month. The same economics are what put more ambitious multi-step setups within reach of smaller teams for the first time.
Starting at all still puts a business ahead of most. The price of trying has rarely been lower. The open question is no longer whether a small company can afford this technology. It is which single job is worth pointing it at first, and who will own the answer.
