OpenAI’s Codex crosses 1 million downloads as AI coding wars heat up

OpenAI

• OpenAI Codex app hits one million downloads in first week on Mac
• Codex usage surged 60 percent week-on-week after debut and GPT-5.3-Codex rollout
• OpenAI may reduce free tier limits as demand and costs rise

Did our AI summary help?

OpenAI has notched up a major early win in the intensifying battle over AI-powered coding tools. CEO Sam Altman confirmed on X that the company’s standalone Codex application has surpassed one million downloads within its first week of availability on Mac, echoing the breakout growth ChatGPT saw after its launch in late 2022.

According to Altman, the surge represents a 60 percent week-on-week increase in Codex usage, following the app’s debut on February 2 and the rollout of the underlying GPT-5.3-Codex model. The numbers suggest strong demand for more autonomous, agent-based coding tools, particularly among developers looking to move beyond basic autocomplete features.

Unlike traditional coding assistants, Codex is positioned as a command centre for what OpenAI calls agentic coding. Powered by GPT-5.3-Codex, the app allows users to run multiple AI agents in parallel, delegate long-running maintenance tasks, and supervise complex projects from a single desktop interface. Altman has claimed that early versions of the model were even used to debug parts of its own training process.

However, the rapid uptake has also forced OpenAI to signal a shift away from generous free access. The one million download milestone was partly driven by a limited-time promotion that opened Codex to ChatGPT Free and Go tier users. That window is now closing.

“We’ll keep Codex available to Free and Go users after this promotion,” Altman wrote on X, “but we may have to reduce limits there.” The move reflects the growing cost of running high-capability models and suggests stricter throttling is likely for lower-priced tiers, while paid subscribers continue to enjoy higher usage caps.

The timing is significant. OpenAI’s momentum comes as rivals gain ground of their own. Anthropic recently said its Claude Code product reached $1 billion in annualised revenue within six months. Meanwhile, tools such as Kilo CLI are pushing a model-agnostic approach, supporting hundreds of AI models across terminals, IDEs and collaboration platforms.

For enterprise leaders, Codex’s early success signals a broader shift in how AI is used in software development. The focus is moving from AI as a passive copilot to AI as an operator capable of managing debugging, testing and deployment workflows. At the same time, the emergence of flexible, vendor-neutral alternatives highlights the risks of ecosystem lock-in.

Codex’s first-week numbers may be impressive, but the next phase of the AI coding wars will likely be decided not by download counts alone, but by how well these tools balance raw capability, cost control and long-term governance.