OpenAI spotlights GPT-6 Astra and its AI strategy

OpenAI presents GPT-6 Astra as the center of its next phase of growth, with advances in web browsing, computer use, programming, cybersecurity, science and professional work. The company says the model expands the range of tasks that businesses and individuals can solve with artificial intelligence.

The announcement is not limited to describing a new model. OpenAI also explains how it wants to turn that capability into a broader business: reach more users, reduce the cost of running AI and use the revenue to fund new models and infrastructure.

From ideas to practical work

OpenAI's thesis is simple: many tasks are not done today because they require too much time, money or specialized knowledge. If AI lowers those barriers, a company can automate processes, serve new markets or test services that were previously not profitable.

One example is the use of agents, systems capable of executing several steps on their own. According to the company, its research teams write code faster, run more experiments and delegate increasingly complex tasks. Within the research organization, agents already generate the equivalent of 3.1 workdays for every human workday.

People still set priorities and review the results. The difference is that they can now spend more time deciding which problems are worth investigating.

OpenAI also says that one of its internal models has found a solution to the Navier-Stokes Millennium Prize problem, an open mathematical question for approximately 90 years. The company presents it as progress in AI's ability to contribute to mathematical research, although the announcement does not replace the independent validation that results of this kind require.

A user base that fuels the business

OpenAI's products reach, according to its own data, more than 1 billion weekly active users and 2.5 million businesses. The company uses that distribution to bring advances in its models to ChatGPT, ChatGPT Work, Codex and applications built through its API.

The company says usage increases over time. In a study of people with individual ChatGPT plans, the daily volume of messages was approximately 50% higher six months after signup than during the first month. Those users had also tried nearly twice as many types of tasks.

OpenAI expects the boundary between personal and professional use to become less clear. Someone who learns to use ChatGPT at home can bring that experience to work, while business tools can change what that person expects from AI in daily life.

Its revenue model combines several channels:

• Free access, partly funded through advertising.
• Subscriptions for users who need more capacity.
• Usage-based payments for businesses and developers.

The logic is that as models improve and cost less, tasks that previously did not justify the investment become viable.

Lower cost per task

To make that growth viable, OpenAI is investing across the technology stack: data centers, chips, software, models and products. Training an advanced model does not require the same resources as running an interactive agent thousands of times, so the company aims to combine in-house components and external providers according to each need.

OpenAI says that GPT-5.6 Sol helped improve its production software and reduced end-to-end execution costs by 20%. Other changes increased the efficiency of token generation, the units of text processed by the model, by more than 15%.

The company also introduced Jalapeño, its first in-house chip for running models. In InferenceX tests with three public models, it delivered between 1.5 and 1.9 times higher peak performance per watt than the commercial systems used for comparison. End-to-end latency, meaning the total time required to obtain a response, was between 1.7 and 3.6 times lower.

OpenAI plans to begin deploying this chip at the end of the year alongside accelerators from NVIDIA, AMD and other partners. These are results from specific tests, not a guarantee that every application will deliver those improvements.

For you, the chip's name will not be the most important effect. What matters is whether a complex task can be completed with fewer attempts, in less time and at a lower price. That could make it profitable to automate processes that previously required specialists or too many hours of work.

OpenAI's strategy depends on those promises becoming real-world usage. The next figure worth watching is not just the power of GPT-6 Astra, but how many new tasks it manages to solve, how much it costs to complete them and how much of that work still requires human oversight.