On March 12, according to "Wired" report,OpenAI is accelerating its AI programming product Codex, trying to catch up with rival Anthropic's Claude Code in the fast-growing AI programming tool market.According to people familiar with the matter, Claude Code currently contributes about one-fifth of Anthropic's revenue, with an annualized scale of more than 2.5 billion US dollars, while Codex's annualized revenue is approximately 1 billion US dollars. This gap means thatIn what is thought to be a multi-trillion-dollar AI programming market, OpenAI is turning from an early leader to a chaser.
OpenAI Codex Product Team
In fact, OpenAI once had a first-mover advantage in this field. In 2021, the company launched the Codex model based on GPT-3 and provided core technical support for Microsoft's code assistant GitHub Copilot. After the product was launched in 2022, it quickly attracted hundreds of thousands of developers to use it.However, with the explosive growth of ChatGPT at the end of 2022, OpenAI has shifted more resources to the research and development of general large models and multi-modal systems. For a time, there was no dedicated team to continue to promote AI programming products.
At the same time, Anthropic chose to focus its resources on strengthening its programming capabilities.The company adds large amounts of real code warehouse data to model training and optimizes it for complex programming tasks. Claude Sonnet 3.5, released in 2024, is generally considered by developers to be outstanding in code generation and debugging, and has also promoted the rapid growth of a number of AI programming tools. The developer platform Cursor has rapidly expanded its user base after integrating this model. The report pointed out that OpenAI tried to acquire Cursor, but the company’s founding team rejected the proposal.
Thibault Sottiaux, Head of Codex at OpenAI
As competitive pressure increases, OpenAI will begin to reintegrate internal resources to promote AI programming product development in 2024.A team led by former DeepMind researcher Thibault Sottiaux worked with product lead Alexander Embiricos and others to develop a prototype of the new AI programming agent. One of the internal projects called "Jam" can run code directly through the command line instead of just generating code snippets, which is regarded as an important turning point for AI programming agents.
Alexander Embiricos, OpenAI Codex Product Lead
In the process, OpenAI is also trying to speed up its catch-up through acquisitions.The company had planned to acquire AI programming startup Windsurf for $3 billion, hoping to gain mature products and an enterprise customer base. However, the deal stalled during cooperation negotiations with Microsoft and was ultimately not concluded. Subsequently, the Windsurf founding team was hired by Google, and the remaining team members were acquired by Cognition, another AI programming company.
With the launch of inference model o1 and subsequent model o3, OpenAI began to strengthen Codex's code generation and automatic debugging capabilities. According to people familiar with the matter, Codex usage was only about 5% of Claude Code in September 2025, but it had risen to about 40% by January 2026, showing a catching-up trend.OpenAI CEO Sam Altman believes that AI programming not only has huge commercial value, but may also become an important path to realize general artificial intelligence (AGI).
The enterprise market has become the core of competition between the two parties. Fidji Simo, head of OpenAI's application business, said the company plans to integrate Codex capabilities into ChatGPT and other products to expand its enterprise user base. Some large companies have begun to encourage employees to use relevant tools. For example, Cisco management publicly stated that employees who do not use AI programming tools may be at a competitive disadvantage in the future.
As the capabilities of AI programming tools continue to increase, discussions surrounding the automation of software development are also heating up in the technology industry.OpenAI co-founder Greg Brockman said that as more and more tasks are completed by AI agents, human engineers may gradually shift to the role of supervising and managing systems rather than writing code themselves.