Frontier Firms Now Generate 8.3x More AI Output Per User Than Everyone Else — Up From 2.6x in January

Frontier firms — the top 10% of companies by monthly AI usage — now generate 8.3 times as many output tokens per active user as typical firms, up sharply from 2.6 times in January. That’s the headline number from OpenAI’s Enterprise Signals report and a companion working paper, “How Organizations Use AI: Evidence from ChatGPT,” published August 12. The gap isn’t closing as adoption spreads — it’s widening, which means the enterprises furthest along are pulling away from everyone else rather than the field converging.

The shift OpenAI documents is from assistance to execution. As of June, Codex generated 64% of combined Codex and ChatGPT output tokens among enterprise customers, evidence that delegated agentic work — not chat-based Q&A — is where enterprise usage is concentrating. That shift isn’t confined to engineering: weekly active enterprise Codex users grew 108x in legal, 41x in sales, 41x in recruiting, and 26x in marketing since February, against 5x in engineering itself. Frontier firms are also far ahead on extensibility — 21% of their weekly active users use Plugins and 19% use skills, versus 9% and 3% at typical firms — and OpenAI notes that internally, 95% of its own employees use Plugins weekly. A case study in the report cites Virgin Atlantic using Codex to refactor legacy code in 30 minutes versus two weeks previously.

That widening curve tracks with what OpenAI documented at the individual-company level two months earlier. When Samsung Electronics rolled ChatGPT Enterprise and Codex out company-wide — to all Device eXperience division employees worldwide — OpenAI reported Codex had passed 5 million weekly users globally, with Korea usage up nearly 800% since February 1. Samsung’s General Manager for OpenAI Korea, Harrison Kim, framed the deployment as treating AI “as a core platform” rather than a departmental tool — the same all-in posture that separates frontier firms from typical ones in the new data.

For consulting engagements, the number that matters isn’t the aggregate adoption rate — it’s the 8.3x. Clients asking whether they’re “behind” on AI are really asking whether the gap is still closeable. OpenAI’s own data says it’s getting harder, not easier, to close.