The enterprise AI usage gap widened to 8.3×

OpenAI’s August Enterprise Signals compares the top tenth of firms by monthly output tokens per active user with firms at the 45th–55th percentiles. The usage ratio rose from 2.6× in January to 8.3× in June.

OpenAI Enterprise Signals, August 12 release. Monthly cohorts ranked by usage; this is not a productivity multiplier.

At frontier firms, 21% of weekly active users used plugins, versus 9% at typical firms. Skills reached 19% versus 3%. These measures suggest deeper use of the tools surrounding the model, rather than simply more chat messages.

The cohorts are ranked again each month, so their membership can change. And 8.3 times the generated output is not evidence of 8.3 times the productivity: tokens can represent useful execution, retries or work that is never accepted.

The companion announcement reports that Codex produced 64% of combined Codex-and-ChatGPT enterprise output tokens in June. It also notes that adopting public companies were already larger and more R&D-intensive than non-adopters. The data describe adoption and use; they do not isolate AI’s causal effect on business performance.