OpenAI data puts EU frontier firms behind global leaders

In Lithuania, Codex and ChatGPT Work generate 85% of the output tokens produced by ChatGPT Enterprise customers. In Slovenia, the same two tools account for 22%, according to a new report from OpenAI Economic Research, which finds that agentic tools already produce more than half of output tokens across its EU sample.

At a glance

• Since May 2026, weekly active users of Codex and ChatGPT Work grew more than eightfold at EU small and medium businesses, which OpenAI defines as having 200 or fewer associated users.
• In September, EU frontier firms, the top 10% by output tokens per active user, generated more than four times as many tokens per user as typical EU firms.
• Every figure describes OpenAI's own ChatGPT Enterprise and Business customers, and the post publishes no absolute user counts, token volumes or sample size behind its percentages and multiples.

If you have not followed Codex, it has had two lives. According to ScriptByAI, OpenAI introduced it on August 10, 2021 as a model that translated plain English into code and powered GitHub Copilot. ScriptByAI says OpenAI retired it in March 2023, revived it in April 2025 as an open-source terminal coding agent and added a cloud agent in May 2025. The same outlet reports that Codex and ChatGPT Work now share models, with GPT-6.1 Sol launching in both on Sep 29, 2026.

Codex and ChatGPT Work produce more than half of output tokens in the EU sample

OpenAI tracks agentic use through two products, Codex and ChatGPT Work, and measures their share of output tokens, meaning the text the models generate in response. Among ChatGPT Enterprise customers in the EU sample, those two tools now account for more than half of all output tokens. In the report's framing, people are no longer just asking questions. They are giving agents tasks to carry from planning through to completion.

That average hides a wide spread. Among ChatGPT Enterprise accounts in the sample, the agentic share ran from 22% in Slovenia to 85% in Lithuania. OpenAI adds that countries differ in the kinds of work people bring to ChatGPT as well as in how heavily they use agents, but the post does not say which tasks those are.

Weekly agentic users at EU small businesses grew more than eightfold since May 2026

The second finding is about smaller firms. OpenAI counts a business as small or medium if it has 200 or fewer associated users. Since May 2026, weekly active users of Codex and ChatGPT Work at those EU businesses have grown more than eightfold, and the report says every member state saw growth.

An earlier OpenAI Economic Research paper, "The Shift to Agentic AI: Evidence from Codex", covered Codex alone and was submitted to arXiv on 25 June 2026. According to the arXiv paper, Codex active users grew more than fivefold in the first half of 2026, and growth was fastest outside the original audience of software developers. The paper also found that Codex use was nearly universal inside OpenAI, while outside adoption was lower and more uneven.

Frontier firms outside the EU generate more than twice the tokens per user of EU leaders

For the third comparison, OpenAI defines frontier firms as the top 10% of firms by output tokens per active user, using only ChatGPT Enterprise and Business accounts. In September, frontier firms in the EU generated more than four times as many output tokens per active user as typical EU firms.

Against the rest of the world, the picture flips. Frontier firms outside the EU generated more than twice as many output tokens per active user as frontier EU firms. EU-based firms made up only 10% of frontier firms globally. OpenAI calls that low, because the EU accounts for a higher share of the world's registered companies.

Sarah Friar, OpenAI's chief financial officer, said there is "still plenty of room for AI use to grow across the EU." OpenAI's own answer to the gap is to help businesses move past experimentation, give workers the skills and confidence to use AI, and make advanced capabilities easier to access and more affordable.

Codex and ChatGPT Work count as agentic because they run tasks in a loop

As Explainx notes, the report uses Codex and ChatGPT Work as its stand-in for agentic use. That is a definitional choice: here, "agentic" means tools that carry out multi-step tasks instead of answering one-off questions. Tech Times describes the loop behind such a system. The agent works out what is needed, calls an external tool to read a file, run code or run a test, gets the result back, updates its picture of the task and picks the next step.

According to Tech Times, the cycle repeats with no human involved until the task is done or the agent hits a limit. Compare asking a colleague a question with handing them a ticket they close on their own. Tech Times reports that OpenAI measured this difference: in the week before June 11, 2026, 60.3% of Codex sessions invoked at least one external tool, compared with 21.9% of ChatGPT sessions.

Every number here comes from OpenAI reporting on its own customers. Without absolute user counts, token volumes, sample size or per-country firm counts, an eightfold rise could rest on a small base. Tech Times also argues that token counts understate agentic work, because handing off an eight-hour task means handing over decisions, not producing more words. In our view, that makes output tokens per user an awkward yardstick for ranking frontier firms in a report about agents.

What the frontier gap leaves open

OpenAI says the next step is turning broader adoption into deeper use. The post sets no target, gives no date for a follow-up report and does not say how far EU frontier firms would need to move to catch up. It also does not explain what kinds of work separate Lithuania from Slovenia. Until OpenAI publishes absolute counts, we cannot tell whether the eightfold small-business growth will hold once the base gets bigger.