According to a CCTV Finance report on June 7, OpenAI is about to conduct its largest upgrade to its chatbot ChatGPT since its launch in 2022, and the new version will be gradually rolled out in the coming weeks.
OpenAI plans to transform ChatGPT into a "super app," integrating programming tools and AI agents, and adding multiple new products to broaden its revenue streams. This upgrade will focus on strengthening its programming product Codex, allocating more resources to it. This means that OpenAI's products will no longer be limited to question-and-answer chatbots, but will primarily feature intelligent agents that can autonomously help users complete various tasks.
Codex is an AI code generation and training model launched by OpenAI. Based on the GPT-3 architecture, it focuses on converting natural language instructions into code in various programming languages. In August 2021, Codex was released as a beta API, and the open-source native tool Codex CLI was launched in April 2025. Last October, the official version of Codex was released, having processed a total of 40T tokens. It added Slack integration, a Codex SDK, and management tools, supporting enterprise-level environment control, analysis, and monitoring, becoming one of OpenAI's fastest-growing products. Currently, it is one of OpenAI's main sources of revenue.
OpenAI is moving forward with its IPO preparations. At the same time, the company faces increasing pressure to boost revenue and pave the way for profitability in preparation for its initial public offering.
According to a previous report by The Information, OpenAI's revenue in the first quarter of this year was approximately $5.7 billion, nearly $1 billion higher than Anthropic's revenue during the same period. OpenAI's growth primarily came from Codex, sales to enterprise clients, and ChatGPT advertising testing.
However, OpenAI's profitability is not optimistic. The company's adjusted operating profit margin in the first quarter was -122%, meaning that after excluding major costs, OpenAI still lost approximately $1.22 for every $1 of revenue. AI model training, inference computing power, product expansion, and user subsidies are still consuming a significant amount of cash. In contrast, Anthropic is expected to achieve an operating profit of approximately $600 million in the second quarter, with a profit margin of about 5.5%, which is close to the break-even point.
Furthermore, OpenAI's user growth has also hit a bottleneck. OpenAI averaged approximately 905 million weekly active users in the first quarter, compared to approximately 920 million in February. Regarding paid users, ChatGPT had 55 million consumer subscribers in the first quarter, up from approximately 47 million at the end of last year.
If a year ago, the core narrative of AI models was who could make the models stronger, this year the market is more focused on who can convert AI usage into revenue, and who can convert that revenue into more stable profits. This means that AI giants must seek out users who can generate high profits, rather than limiting themselves to free chat clients.
Take Anthropic as an example. As previously reported, in an effort to attract Wall Street clients, Anthropic PBC released a series of new AI agents designed to handle a wider range of financial services tasks. The company's AI agents can be used to draft client meeting presentations, review financial statements, and escalate cases to compliance reviews.
Anthropic stated that it will enable the Claude model to work better with third-party software such as Excel, PowerPoint, and Outlook, and integrate data from financial industry partners such as Dun & Bradstreet and Moody's. Anthropic is also deepening its presence in the financial industry through new joint ventures with Blackstone Group, Hellman & Friedman, and Goldman Sachs. These joint ventures will drive the adoption of its software in more enterprises.
Google is also applying its AI agents to more scenarios. At the Google I/O 2026 developer conference in May, Google announced that it is fully integrating its AI agents into all core entry points. From the search box to the Chrome browser, from Android phones to smart glasses, Gemini is no longer just a conversational assistant, but a continuously running AI agent that can perform tasks across applications: it can track information, generate content, call tools, and even directly complete orders and operation processes for users.
At the aforementioned developer conference, Google CEO Sundar Pichai stated that the past 12 months have been a year of "rapid growth" for Google. Google processed 3.2 quadrillion tokens monthly, a sevenfold increase year-over-year; and the Gemini app boasted 900 million monthly active users.
Like Anthropic and Google, OpenAI also needs to find new profit growth points. As its biggest redesign ever, this significant investment in Codex raises questions about whether it will create the expected value.
This article is from the WeChat official account "Jiemian News" , authored by Hou Ruining, and published with authorization from 36Kr.
