OpenAI said active users of its artificial intelligence (AI) models have surpassed 1 billion and the number of adopting companies has topped 2 million.

IT outlet ITMedia reported on Aug. 2 local time that the figures were disclosed by Chief Financial Officer Sarah Fryer (사라 프라이어) in an official blog post titled "Building Abundant Intelligence".

The announcement comes a day after OpenAI cut prices for "GPT-5.6 Luna" and "GPT-5.6 Terra", models in the "GPT-5.6" family. OpenAI presented a structure in which cost reductions and broader use feed back into research and infrastructure investment, without treating user growth and price cuts as separate issues.

OpenAI said usage intensity is also rising. Users who have been signed up for six months send about 50 percent more messages per day on average, and the types of work using ChatGPT have roughly doubled. OpenAI said agent-style work via Codex accounts for 99.8 percent of weekly output tokens internally. This is interpreted as meaning the share of actual work automation is increasing compared with simple conversational use.

Fryer stressed optimisation of surrounding systems, rather than changes to the model itself, as the backdrop to the price cuts. She said improving inference retention and context management alone can deliver major efficiency gains, citing a case in which Sol's score on the ARC-AGI-3 benchmark rose to 38.3 percent from 13.3 percent while output tokens fell to one-sixth. She said performance and costs can be improved together by overhauling systems without changing the model.

OpenAI also disclosed achievements in reducing operating costs. It said the engineering team worked with "GPT-5.6 Sol" to optimise software in the real operating environment, cutting the end-to-end cost of providing the model by 20 percent. It also said improvements to speculative decoding lifted token generation efficiency by more than 15 percent. OpenAI described the price cuts as a step to return these operating efficiency gains to customers.

Fryer stressed that the lower the cost, the wider the range of work that can deploy AI. She said that as the cost of useful intelligence falls, the number of tasks worth trying increases, and broader use generates real data on revenue and demand that can be reinvested in next-generation research and infrastructure. OpenAI's decision to disclose growth metrics and pricing policy together is also seen as highlighting this virtuous cycle.

Fryer also presented a new standard for performance metrics. She said what customers want is not tokens themselves but outcomes such as resolving support tasks or launching software. She said an appropriate metric is the cost per successful outcome, including retries and human intervention. That means putting the cost of completing real work at the centre rather than simple usage volume.

In this context, OpenAI's announcement did not stop at disclosing user counts and the scale of corporate adoption. It emphasised that the price cuts are not simple promotion but the result of improved operational efficiency, and that the efficiency feeds back into broader demand and reinvestment. A key focus is likely to be how much agent-style use grows in corporate settings and whether cost reductions lead to additional pricing policy.