What an 80% AI Price Drop in Three Weeks Actually Means
OpenAI slashed the price of its budget AI model by 80% in just three weeks — pushed by Chinese competition — and that changes which high-volume tasks are worth building.
- 80%
- price cut for Luna in three weeks
- $0.20
- Luna's new cost per million input tokens
- 46%
- enterprise AI traffic on OpenRouter going to Chinese models
- 3 weeks
- from GPT-5.6 launch to major repricing
Three weeks after OpenAI launched its new GPT-5.6 model family, it cut the price of the budget model — Luna — by 80 percent. Prices do fall in tech, but this fast, this sharply, tells you something: the race for AI is now a price war, not a performance contest.
Tokens are the small units of text an AI reads and writes. Luna used to cost $1 for every million input tokens. Now it costs 20 cents. Send a million messages through the API each day and your monthly bill just fell by tens of thousands of dollars. Tasks that seemed too expensive to automate a month ago are not anymore.
OpenAI did not do this out of generosity. Data from OpenRouter — a platform that routes AI requests to different providers — showed Chinese models handling nearly half of all enterprise AI traffic there. DeepSeek, the biggest Chinese competitor, charges a fraction of US prices. OpenAI had to respond, and fast.
There is an interesting detail in how the cut was funded. OpenAI says its flagship model, Sol, helped find the efficiency gains — by rewriting its own production code — reducing the cost of running itself by around 20 percent. AI is now helping cut the price of AI.
For anyone building products, the practical takeaway is simple: if you shelved a high-volume AI task because the numbers did not work, run them again. Summarizing emails, screening applications, moderating content — these need millions of AI calls a month. At 20 cents per million tokens, the math on most of those tasks now works.