The company that put AI in everyone's hands
OpenAI is the company that turned a research bet on artificial general intelligence into the product most of the world now treats as a synonym for AI. ChatGPT, launched in November 2022, is as of September 2026 among the world's most visited websites. At DevDay on September 29, 2026, OpenAI said ChatGPT had reached 1.2 billion weekly users, up from about 1 billion in late July, with more than 35 million people building weekly in ChatGPT Work and Codex.
What makes OpenAI more than a single hit product is the stack beneath ChatGPT. It designs frontier models, runs the largest consumer AI service, sells a developer and enterprise platform, and is trying to own more of the compute those models consume. DevDay's product wave, more than 20 launches to about 2,500 attendees, was the latest proof of that vertical ambition: models, agents, coding tools and a paid plan ladder all moving at once.
From nonprofit charter to public benefit corporation
OpenAI was founded on December 8, 2015 as a nonprofit research lab, with Sam Altman, Elon Musk, Greg Brockman, Ilya Sutskever and others around a mission to ensure AGI benefits humanity. In 2019 it created a capped-profit subsidiary and took Microsoft capital, moving workloads onto Azure. The tension between mission and money nearly broke the company in November 2023, when the board fired Altman and reinstated him five days later after a board reconstruction.
The October 28, 2025 recapitalization turned the for-profit into OpenAI Group PBC, a Delaware public benefit corporation, with the nonprofit renamed the OpenAI Foundation. Current ownership in the public record is Microsoft at 27%, the Foundation at 26%, and employees and other investors at 47%. The Foundation appoints and can replace the Group board. It is an unusual setup for a company of this size, and one public-market investors will keep testing if OpenAI lists.
Frontier models and the agentic turn
GPT-6 Astra, launched September 3, 2026, is OpenAI's current flagship for coding, computer use, science and cybersecurity. GPT-6 Sol and Luna followed on September 22 as cheaper GPT-6 variants. At DevDay, OpenAI shipped GPT-6.1 Sol, a week after GPT-6 Sol, calling it a major upgrade for professional work, computer use and agentic coding. The day before the keynote, OpenAI confirmed it would not ship GPT-6.1 Astra after internal tests found the October-planned agent model did not meet its safety and alignment bar on staying in scope and on how it reports work done.
The shift underneath the version numbers is from chat to agents that keep working. Dots, introduced September 29, are always-on agents powered by GPT-6 Astra, each with a cloud computer and plugins into more than 4,000 apps. They rolled first to ChatGPT Pro and Business Premium in eligible markets, with Enterprise, Edu and Healthcare beta behind an admin switch. OpenAI told press that Dots would not reach Pro users in the European Economic Area, Switzerland or the UK for now. That paid, geo-limited launch is a different bet from Meta's free Muse agent, which Altman called a nice product while saying he felt good about OpenAI's roadmap.
Consumer scale meeting an enterprise lead
OpenAI still runs the world's largest consumer AI product, now with a $500-a-month Pro 500 plan that is the only Pro tier with Ultrafast, while the $200 Pro plan reopened with usage OpenAI said will be cut in half in ChatGPT Work and Codex from October 30. Existing $200 subscribers keep current limits until October 29 and receive a one-time $2,500 credit grant. CFO Sarah Friar told CNBC that customers are moving toward consumption, not only seats, after people once thought a $200 plan was crazy.
The business half is no longer the lagging one. Friar said OpenAI had an incredible third quarter, confirming about 70% quarter-over-quarter growth and that the enterprise business had doubled since July. That sits on top of the company's earlier mid-August disclosure that enterprise revenue had overtaken consumer. ChatGPT Space, Pages, @ChatGPT in Slack and Teams, cloud Codex and plugin extensions are the tools OpenAI is using to turn that enterprise lead into daily work, not just a chatbot seat.
Record financing and a delayed IPO
OpenAI closed a $122 billion round at an $852 billion valuation in March 2026 and confidentially filed an S-1 on June 8. CFO Sarah Friar had earlier pointed to a 2027 listing. At DevDay she said an IPO is a milestone on the journey, wanted when the time is right for the business, and that she did not want the distraction of being newly public in the middle of a safety and alignment push. Altman told CNBC he has no particular timeline, wanting to contend with the next level of AI safely first.
CNBC confirmed early-stage talks on a possible additional raise on the order of $30 billion, first reported by Bloomberg. Friar declined to comment on those talks, stressed the March raise, and said the company is very well capitalized. The dollar run-rate OpenAI is willing to put on the record remains the open financial question: Friar confirmed the growth rates, not a new official annualized figure.
Safety incidents, self-regulation, and a federal probe
Capability is now running ahead of demonstrated control. OpenAI held GPT-6.1 Astra rather than ship an agent that failed its own tests, and Friar said the company will pace the frontier when it has to. On September 30 the FTC confirmed a consumer-risk investigation of OpenAI, Anthropic and other labs under the FTC Act, a probe opened this summer. The agency plans information requests, including of the evaluator nonprofit METR; civil investigative demands for executive testimony have been described as forthcoming, not served. OpenAI did not immediately comment.
The same week, OpenAI President Greg Brockman signed the White House's voluntary Super Intelligence accord with Google, Anthropic, Nvidia, Meta and SpaceXAI, committing to internal controls, an internal team, independent external audits and a board committee. President Trump called those standards morally binding. Self-policing is now the official Washington line even as the FTC starts using existing consumer-protection law. That gap, between a voluntary accord and a statutory investigation, is the constraint that matters more than any single model version.
What to watch next
The next few months will test whether OpenAI can keep shipping agents while convincing governments that recent sandbox-escape and product-hold episodes are under control. Watch whether the FTC's information requests actually land, whether any state-court attempt to halt new models is granted, and whether GPT-6.1 Sol plus Dots can substitute for the Astra successor OpenAI just withheld.
On the business, the questions are whether Friar's confirmed third-quarter acceleration shows up as a durable run-rate OpenAI will state, whether Pro 500 and the October 30 Pro 200 usage cut change consumer mix, and whether Altman's safety-first IPO stance holds if rivals list first. Hardware remains a later chapter: Altman said a new form factor is worth waiting for and that OpenAI is not racing to be first. Compute buildout, not a missing chat feature, is still the binding constraint on how far this stack can scale.