The numbers OpenAI did not give
OpenAI said weekly active Codex users in Germany have grown 20-fold since January 2026. In June, it had reported growth of more than sevenfold. The company again withheld the underlying user count, making the growth rate hard to assess in absolute terms.
Dots remains unavailable to personal Pro accounts in the European Economic Area, Switzerland and the United Kingdom. Some business accounts already have access. OpenAI’s Romain Huet, head of developer experience, said the assistant should become more widely available in Europe “within a few weeks,” following talks with regulators.
Nearly one in five desktop users turns to Codex skills for documents, presentations and spreadsheets. That suggests the coding tool is finding uses beyond software development, though OpenAI offered no further detail on how people use those features.
Compute follows the agents
At the event, OpenAI executives said customers had rarely asked for a model that understands video sequences like Google’s Gemini. The company has not prioritised that capability; for now, its model can analyse individual frames.
Participants also revisited Sora, OpenAI’s video generator, which closed in March. They cited the high resource costs and said customers mainly want models and agents for coding. OpenAI did not rule out bringing Sora back.
The pressure is not limited to video. Agents run on their own virtual computers, and the amount of AI-generated code is rising quickly, increasing demand for conventional processor capacity. OpenAI has formed a team to review the company’s internal codebases every week in an effort to reduce CPU load.
The platform is the bigger story
OpenAI is also assembling the infrastructure and distribution for developers to build on its ecosystem. With its Agents API, the company provides infrastructure for agents; plugins and web apps offer developers more ways to put products on the platform.
The pitch is less technical work and access to an audience OpenAI estimates at 1.2 billion weekly ChatGPT users. The model resembles early app stores: developers make the products, while OpenAI supplies infrastructure and a route to customers.
Codex Harness, the system’s core runtime, is open source, so developers can inspect the code and run it themselves. But access to managed services, platform integrations and customer discovery through ChatGPT could make moving elsewhere harder for businesses built around those services.
My read is that open-sourcing the runtime and owning the distribution are compatible strategies. Developers can retain control over the code while becoming more dependent on the services around it. The announcement says less about how much demand developers will actually find through ChatGPT than about how much of their business OpenAI hopes to host.
Daily AI news
Every day we pick what actually matters in AI and explain it plainly — no hype, no filler. Subscribe if you want to follow where the industry is going.
Only what matters — every day
Follow on X