OpenAI says agentic tools now exceed half of EU enterprise output tokens

Agentic products are no longer a side lane in OpenAI’s European enterprise numbers. The company says Codex and ChatGPT Work now account for more than half of output tokens in its EU sample. OpenAI published that finding on October 14, 2026, in “The EU’s next chapter in AI is agentic,” with a companion PDF, “Europe’s Agentic Opportunity,” dated October 2026. The note calls the work the latest report from OpenAI Economic Research. It is OpenAI’s own usage analysis, not an independent audit and not a product launch.

What the sample actually counts

The agentic bucket is Codex plus ChatGPT Work, not every ChatGPT message. The blog says that among ChatGPT Enterprise customers, agentic use already represents a substantial share of activity. In the EU sample, Codex and ChatGPT Work now account for more than half of output tokens. The PDF’s chart is labeled more tightly, and its account note is slightly broader. It shows a 28-day trailing share of enterprise-generated text tokens, excluding a category called Other, and the chart note says the analysis covers ChatGPT Enterprise and Business accounts.

ChatGPT Work launched in July 2026. On that chart it does not appear through July. It is 3 percent in August, 12 percent in September, and 21 percent on September 30. Codex moves from 20 percent in May to 34 percent on September 30. Those two shares are 55 percent of the categories shown; ChatGPT is the other 45 percent. The PDF’s prose says that by the end of September, after ChatGPT Work’s launch, agents via Codex and ChatGPT Work accounted for more than 50 percent of output tokens generated by firms in the EU. Anyone lining the blog’s sentence up against the chart should keep the “excluding Other” note in view. The blog does not spell out that exclusion.

Member states are not on the same curve

The EU figure is not the national one. Among ChatGPT Enterprise accounts in the blog’s sample, the share of output tokens from Codex and ChatGPT Work ranged from 22 percent in Slovenia to 85 percent in Lithuania. The PDF’s September 2026 country table, noted as an analysis of ChatGPT Enterprise and Business accounts, has the same endpoints. Cyprus is at 81 percent, Hungary at 79 percent, and Poland at 76 percent. Germany is at 51 percent and France at 55 percent. Spain is at 30 percent and Austria at 26 percent. OpenAI says smaller-population countries, including Cyprus, Estonia, and Lithuania, show relatively high agentic use within their larger firms. It also says the kinds of work people bring to ChatGPT vary by country. These are shares inside the sampled accounts, not a census of European companies.

This analysis defines a small or medium business as one with 200 or fewer associated users. Weekly active users of Codex and ChatGPT Work among EU SMBs were 8.1 times the May 1–7 level by October 5, with growth in every member state. The series was 1.2 times in June and 1.25 times in July, then 3.1 times in August, 6 times in September, and 8.1 times on October 5. The PDF says a major part of that growth followed the introduction of ChatGPT Work. From September 29 to October 5, every EU country was more than five times its May level. Belgium grew fastest, at 12.1 times, followed by Italy at 10.8 times and Luxembourg at 10.0 times. Finland was the low end of that table, at 5.1 times.

Intensive users are ahead, and still behind the global frontier

Frontier firms are the top 10 percent of firms by output tokens per active user. That comparison includes only ChatGPT Enterprise and Business accounts. In September, frontier EU firms generated more than four times as many output tokens per active user as typical EU firms. Relative to typical firms in May, the PDF puts EU frontier firms at 6.1 times and typical EU firms at 1.5 times. Frontier firms outside the EU are at 12.9 times, more than twice the EU frontier. The blog prints the same September multiples: 12.9 times, 6.1 times, 1.6 times, and 1.5 times for the global frontier excluding the EU, the EU frontier, typical firms outside the EU, and typical EU firms. In May, the PDF had the EU frontier at 3.9 times and the global frontier at 5.4 times. The PDF says the gap between typical firms and top firms widened as agentic use increased.

EU-based firms were 10 percent of frontier firms in that September sample. The blog calls that share low relative to the EU’s weight among registered companies worldwide. In the PDF’s geography split, North America accounted for roughly 65 percent, APAC 13 percent, and other EMEA countries 11 percent. South America was 1 percent, and unmapped firms were under 1 percent.

Sarah Friar, OpenAI’s chief financial officer, described the shift more broadly than the token tables do. “More than half of the AI use among our EU enterprise and business customers now involves these tools, and the number of small business users has grown 8x across the region,” she said. “AI use” is not the same measure as output tokens, and 8x is the rounded companion to the 8.1 times SMB series. She said there is still plenty of room for AI use to grow across the EU, and that giving people the tools, skills and confidence to use AI well will help build stronger businesses and growing economies, with benefits felt across Europe.

The report says the EU’s opportunity is to turn broader adoption into deeper use. It separates that intensity gap from a wider adoption figure it states on its own: 20 percent of EU companies with at least 10 people employed used AI in 2025, up from 13.5 percent in 2024. The token figures do not say whether heavier use changed revenue, headcount, or hours worked.

Start with the October 14 note, then open the PDF and check the September 30 row — Codex at 34 percent, ChatGPT Work at 21 percent — against the blog’s claim that those two products are now more than half of output tokens in the EU sample.

Sources

• The EU’s next chapter in AI is agentic
• Europe’s Agentic Opportunity