OpenAI surpasses the threshold of one billion active users

OpenAI has surpassed the symbolic threshold of one billion active users, less than four years after the launch of ChatGPT in November 2022. The company also has more than two million companies using its models and products.

“Our models now reach more than one billion active users and more than two million businesses. As people gain confidence in the technology, they use it more deeply,” OpenAI said in a insights into the efficiency of the new GPT-5.6 family.

The growth of ChatGPT

OpenAI was founded in 2015, but launched its first product intended for the general public only in November 2022. In the first months of 2026 the company declared over 900 million weekly active users on ChatGPT and more than 50 million consumer subscribers. The billion figure includes all of the company’s models and products, with its best-known model in the foreground, but also Codex, dedicated to programming assisted by artificial intelligence, the APIs used by developers and the tools intended for businesses. The spread of ChatGPT has become progressively more global. According to data published by OpenAI, more than half of active users predominantly use languages other than English, while the fastest relative growth was recorded in Africa and Asia.

Over two million client companies

The corporate market has become one of the group’s priorities. More than two million enterprises use OpenAI models to develop products, automate processes, and manage tasks in engineering, customer service, finance, sales, and operations. OpenAI also reported in recent months that over nine million paid business users were using ChatGPT for work. During the same period, Codex had reached 1.6 million weekly users. The growth in the professional use of artificial intelligence is linked above all to the spread of agents, systems capable of carrying out complex tasks, using external tools, evaluating the results and autonomously correcting their work.

The challenge of computing costs

The expansion of agents, however, has rapidly increased the demand for computing power and the costs incurred by businesses and developers. For this reason, part of the market is moving towards smaller and cheaper models, including open-weight ones developed in China. OpenAI claims to have addressed the problem by intervening not only on the models, but on all the main levels of its infrastructure: training, system execution and the mechanism that coordinates the agents and the use of the tools. The company says it has trained GPT-5.6 to complete tasks by taking more direct paths and using fewer tokens. The interventions also concerned the balancing of workloads, the management of temporary memory, the optimization of the software used by the processors and the reduction of repeated operations.

Artificial intelligence used to improve models

According to OpenAI, GPT-5.6 Sol directly contributed to some of the optimizations necessary for its operation. The model was used to write and verify parts of the software that handles the calculations on the chips and to design a smaller system tasked with anticipating the responses of the main model.

This technique, called speculative decoding, allows a lighter model to propose a sequence of tokens in advance, which the main system can verify simultaneously. When proposals are accepted, more text can be generated with fewer computation steps. OpenAI also claims to have improved context management in agent systems, reducing the accumulation of useless information, the inefficient use of tools and the repetition of activities already carried out.

The three models of the GPT-5.6 family

The GPT-5.6 family, presented at the beginning of July, consists of three models. Sol is the flagship system, intended for the most complex activities; Terra is the intermediate model, designed for daily work; Luna is the fastest and most economical, designed for high-volume activities.

According to OpenAI, Sol with the maximum reasoning level outperforms Claude Fable 5 in the Artificial Analysis Coding Agent Index at less than half the cost. Terra would achieve performance comparable to GPT-5.5 at half the price, while Luna was designed to maximize economic efficiency.

GPT-5.6 is available on ChatGPT, Codex and through the OpenAI API. Free and Go users can access Terra, while Plus, Pro, Business and Enterprise subscribers can use Terra and Luna. Sol is available in the plans and methods foreseen for the most advanced activities.

80% cut in the price of Luna

OpenAI announced an 80% cut to the price of GPT-5.6 Luna and 20% to that of Terra. The cost of Sol remained unchanged. In the API, Luna now costs $0.20 per million incoming tokens and $1.20 per million outgoing tokens. Terra instead costs 2 dollars per million incoming tokens and 12 dollars for outgoing ones. ChatGPT and Codex subscription prices and usage limits do not change. However, the use of Terra and Luna will require a lower number of credits, due to the reduction in costs of the two models.

OpenAI has also introduced a faster mode in the API for GPT-5.6 Sol, intended for tasks where response time is a priority. The feature offers faster speed than standard processing, without changing the capabilities of the model, but at double the price. The strategy aims to strengthen OpenAI’s position in a market where competition is increasingly based on the relationship between performance and costs. With one billion users and a growing enterprise customer base, the company is now looking to make advanced models usable on a larger scale, reducing the cost required for each business.