OpenAI and Anthropic spent June fighting on price. OpenAI's new Luna model runs at about one dollar per million input tokens and six dollars per million output tokens. Anthropic, under pressure, dropped a planned charge on its developer tools rather than risk pushing builders to its rival. When two companies this size compete on price, the people paying the bills come out ahead.
What a Token Actually Costs You
Tokens are how AI usage is measured. Roughly, a token is three quarters of a word, so a million tokens is around 750,000 words. At Luna's rates, processing that much text costs a few dollars. Two years ago the same work cost many times more. The same AI feature gets cheaper to run every few months without you changing a thing.
Why This Matters for Smaller Businesses
Price is what decides whether an AI feature is worth building for a small business. A chatbot that costs more to run than the staff time it saves is a toy. The same chatbot at a tenth of the cost is a tool. Falling prices move more ideas across that line. Things that did not make financial sense for a Gauteng SME last year, automating quotes, screening enquiries, drafting every first reply, are starting to.
How to Take Advantage
Build now, on whatever model is affordable today, and keep your setup simple to switch. The businesses that already run an AI feature ride the price drops down automatically: they swap to the cheaper model when it lands and keep the saving. The cost of waiting is not just the months without the feature, it is starting the build from scratch when everyone else is already three versions in.