OpenAI says the EU has moved past asking AI questions and started handing it work. In a report published October 14, 2026, Europe’s Agentic Opportunity, OpenAI Economic Research finds that its agent products, Codex and ChatGPT Work, produced more than half of the output tokens at its EU business accounts by the end of September, and that weekly active users of those tools at EU small and medium businesses grew 8.1x since May.
Every number in it comes from OpenAI’s own customers. That makes it a good read on how fast agents are spreading inside one vendor’s base, and a poor read on Europe as a whole. Here is what the report shows, country by country, and what it does not.
On this page · 6 sectionsOpen
OpenAI says Codex and ChatGPT Work produced 55% of output tokens at EU ChatGPT Enterprise and Business accounts by September 30, 2026.
Weekly active users of those agent tools at EU small and medium businesses, defined as 200 or fewer users, grew 8.1x from early May to early October.
The agent share varies widely by country: 85% in Lithuania, 51% in Germany, 30% in Spain and 22% in Slovenia.
Frontier EU firms produce over four times the output tokens per user of typical EU firms, yet frontier firms elsewhere produce over twice as much again.
EU firms are 10% of OpenAI’s global frontier firms, against 65% for North America.
§ 01What the report measures
The announcement and the 10-page PDF cover three things: how much of the work on ChatGPT Enterprise and Business accounts now runs through agents, how fast small businesses are adopting those agents, and how far the heaviest EU users sit from the heaviest users elsewhere.
| Measure | Figure | What it covers |
|---|---|---|
| Agentic share of output tokens, EU firms | 55% (September 30) | Codex and ChatGPT Work on EU ChatGPT Enterprise and Business accounts, 28-day window |
| Growth in agentic weekly active users, EU SMBs | 8.1x | September 29 to October 5 against May 1 to 7; firms with 200 or fewer users |
| Every member state, SMB growth | above 5x | Belgium fastest at 12.1x, Finland slowest at 5.1x |
| Frontier vs typical EU firms | over 4x | Output tokens per active user, September |
| EU share of global frontier firms | 10% | North America 65%, APAC 13% |
| EU companies with 10+ staff using AI | 20% in 2025 | Up from 13.5% in 2024, cited by OpenAI as background |
The last row is the useful one for context. Broad adoption across EU companies is still a minority; the 55% describes depth of use inside companies that already pay OpenAI.
§ 02Country by country
The agent share of output tokens differs sharply between member states. Lithuania leads at 85%; Slovenia trails at 22%. The report notes that “countries with smaller populations like Cyprus, Estonia, and Lithuania have relatively high agentic use within their larger firms.”
Small businesses tell a different story. Countries with low agent shares at large firms, such as Slovenia (9.2x) and Spain (8.8x), grew faster among small businesses than leaders like Lithuania (6.8x).
The report also maps the most distinctive work task per country from a random sample of messages, but that analysis uses Free, Go, Plus and Pro plans, not business accounts. Germany’s most distinctive task is teaching; Austria’s is legal research; Sweden’s is advising on technology.
§ 03The frontier gap
OpenAI’s real argument is about depth. Frontier firms, the top 10% by output tokens per active user, pulled away from typical firms between May and September, and the EU’s frontier pulled away less than everyone else’s.
Typical firms barely moved on both sides of the Atlantic. The gap is at the top: EU frontier firms reached about half the intensity of frontier firms outside the EU, and the EU holds only 10% of the frontier firms. OpenAI’s CFO Sarah Friar framed it this way: “Yet there is still plenty of room for AI use to grow across the EU.”
§ 04What the report leaves out
| The report shows | It does not show |
|---|---|
| Agent share of output tokens on OpenAI business accounts | Agent adoption across all EU businesses |
| Growth in weekly active users of two OpenAI products | Sample sizes or how many firms are counted |
| Growth since May, before ChatGPT Work launched in July | Growth with the launch effect removed |
| Distinctive work tasks per country | Business-account tasks (that analysis uses consumer plans) |
On the launch effect, the report is candid: “A major part of this growth followed the introduction of ChatGPT Work.” An 8.1x jump from a baseline set before a product existed is partly a product launch, not only a change in how businesses work.
Spanish coverage in Europa Press carried the country figures the same day, and OpenAI publishes the underlying charts on its Signals research page.
§ 05How it fits with OpenAI’s US report
Two weeks earlier, OpenAI’s small business report found agentic tools made two-thirds of small-business output tokens in August, which we read in our record of that report. The EU report uses a different cut (all business accounts for the 55%, firms with 200 or fewer users for the growth figure), so the two are not directly comparable. Both point the same way: on OpenAI’s paid plans, most output now comes from tools that carry out tasks rather than answer questions.
For EU readers, this lands the same day the AI Office sent its first requests for information to OpenAI and Anthropic, which we covered in our EU AI Office record.
§ 06Sources
- OpenAI: The EU’s next chapter in AI is agentic (October 14, 2026)
- OpenAI Economic Research: Europe’s Agentic Opportunity, PDF (October 2026)
- OpenAI: Helping small businesses put AI to work (September 30, 2026)
- Europa Press coverage (October 14, 2026)
- OpenAI Signals
Q1Does 55% mean most EU businesses use AI agents?
No. It is the share of output tokens at OpenAI’s EU ChatGPT Enterprise and Business accounts that came from Codex and ChatGPT Work. The report itself cites a much lower adoption figure: 20% of EU companies with at least 10 employees used AI in 2025, up from 13.5% in 2024.
Q2How does OpenAI define a small or medium business?
A business with 200 or fewer associated users on its plans. The 8.1x compares weekly active users of ChatGPT Work or Codex in September 29 to October 5 with May 1 to 7.
Q3Which countries grew fastest?
Among small and medium businesses, Belgium (12.1x), Italy (10.8x) and Luxembourg (10.0x). Finland grew slowest at 5.1x, and every member state passed 5x.
Q4What is a frontier firm here?
The top 10% of firms by output tokens per active user, measured on ChatGPT Enterprise and Business accounts. EU frontier firms reached 6.1x the May level of typical firms; frontier firms outside the EU reached 12.9x.
Q5Does this cover CellCog?
No. The data covers OpenAI’s own products. CellCog’s AI employees run on Claude Opus 5.5, and CellCog is not part of the report.