Four months ago, OpenAI was still being judged against a missed target: ChatGPT had come close to 900 million weekly active users but had not reached one billion. On July 31, CFO Sarah Friar used a broader company post to say OpenAI's models now reach more than one billion active users and more than two million businesses.
This was not a new model launch. It read more like an operating memo for customers, investors and compute partners: usage is large, API prices are moving lower, and inference efficiency must keep improving.
After scale, prices move first
OpenAI linked the milestone to the July 30 GPT-5.6 price cut. Luna now costs $0.20 per million input tokens and $1.20 per million output tokens, down 80%. Terra now costs $2 and $12, down 20%. Sol pricing is unchanged, while Fast mode promises up to 2.5x standard processing speed at twice the price.
The Wall Street Journal separately reported the same user and business counts and the Luna/Terra cuts. That matters because the user milestone is being presented with a cost reset, not as a simple victory lap.
“Customers do not buy tokens for their own sake.”
Friar's line reframes the API bill. Buyers care less about the token unit and more about the cost of a solved support case, shipped code, reviewed contract or completed research task.
The bill grows with usage depth
OpenAI says people send roughly 50% more messages per day six months after signing up and use ChatGPT for about twice as many kinds of work. Internally, agentic work through Codex now accounts for 99.8% of OpenAI's weekly output tokens.
The Chinese market angle is straightforward: one billion users make every routing, caching and context-management decision financially large. Model pricing is only the visible layer; the real bill depends on how many turns a task takes and whether the system avoids repeated work.
Inference efficiency becomes the growth condition
OpenAI's engineering post says GPT-5.6 Sol helped reduce end-to-end serving costs by 20% and improved token-generation efficiency by more than 15%. Its ARC-AGI-3 write-up gives the system-level version: the same model moved from 13.3% to 38.3% RHAE after retained reasoning and compaction, while using about one-sixth as many output tokens.
The next test is demand elasticity. Lower Luna and Terra prices should expand API use; Sol Fast should show up in high-value enterprise work; Codex should keep moving beyond engineering teams. If those numbers follow, the one-billion-user claim becomes more than traffic. It becomes the base for an AI business that can pay for its infrastructure.
Sources: OpenAI company post, OpenAI GPT-5.6 pricing announcement, OpenAI GPT-5.6 engineering post, CocoLoop, OpenAI ARC-AGI-3 research note and The Wall Street Journal; sources verify user scale, business count, API pricing, serving-cost reduction, token-generation efficiency, ARC-AGI-3 RHAE and output-token scope.