Codex Reaches 4M Monthly Active Users (5X Growth Over The Last 3 Months)

Codex, the AI system developed by OpenAI that can write, debug, and even evaluate code, has now achieved 4 million weekly active users, growing 5x in the last 3 months and at a monthly pace of 70%. That’s according to a recent interview with Arnaud Fournier, CTO of DeployCo (the enterprise division of OpenAI). “The price of intelligence has decreased by a factor of 100 in the last 18 months,” he said.

And indeed, overall, AI has gotten much cheaper to use in 2024. However, there’s a twist. According to Fournier, OpenAI’s newest premium model, GPT-5.5, is 49–92% more expensive than the previous one. Why? Because it takes longer to come up with answers now. Instead of giving quick responses, it thinks harder about the question. In other words, AI is becoming cheaper for most things, but at the same time, it’s getting more expensive for situations when you really need deep thinking.

1Where Codex is actually being used

Here are three case studies of enterprises that have adopted AI at scale, according to Fournier. BBVA Bank moved from automating credit documents once per year to evaluating risks daily and weekly. Same thing, different frequency. John Deere cuts down the use of chemical fertilizers in fields, saving money and helping the planet. And Stadler, a German waste sorting company with around 650 employees, says that 85% of its workers use ChatGPT every day. Not just playing around, but using it as a tool.

In Germany, Codex saw a 720% increase in adoption since January 2026. That makes it the number one country in Europe and one of the top 5 globally. One of the reasons for that might be the engineering mindset and the desire for hard metrics in Germany, according to Fournier. Not exactly the first place you’d expect to find widespread AI adoption, but the numbers speak for themselves.

2The ROI question nobody has fully answered yet

What kind of ROI do companies get from AI? “It’s still early,” Fournier said. Most deployments happened within the last 6–12 months, which is not enough time to accurately calculate the overall ROI. His recommendation? “Just buy a Codex license for $20 and try it out yourself.” Don’t rely on other people’s experience. Try it on your own project.

That’s great advice, but it also tells us something about the current state of AI adoption. If Codex has 4 million weekly users and grows 70% per month, then it’s not an edge case. It’s mainstream. And if we’re still trying to figure out the ROI, then it’s not obviously positive. Those who see the biggest benefits are those who got in early and created dependencies (like BBVA or John Deere) rather than those who run a few tests on low-impact projects.

Finally, Fournier brushed off worries about the EU AI Act, saying that it’s “barely a brake” in reality. That’s interesting coming from someone who deploys AI solutions in European companies. If the EU AI Act was supposed to slow down enterprise AI deployments, then it doesn’t seem to be having much effect on the 720% growth rate in Germany.