On July 14, Codex engineering lead Tibo Sottiaux posted on X: “Hello. We have reached 8M active users across Codex and ChatGPT Work. We are once again resetting the usage limits for all. And we continue to not have the 5h rate limit as well, allowing everyone to explore the boundaries of GPT-5.6 Sol and discover how ambitious you can be.” That is the whole primary source — one post from a named OpenAI engineer, not an official announcement page.
It is a real number with a real name attached, and it is the milestone the five-hour-cap story said it would wait for. But before anyone reads it as Codex going from 5 million to 8 million in six weeks, look at what is actually being counted.
The number counts a different thing than the last one
The 5 million figure OpenAI published June 2 was weekly users of Codex. The 8 million figure is active users across Codex and ChatGPT Work, combined. Two of the three words changed — “weekly” became “active,” and “Codex” became “Codex and ChatGPT Work.”
That matters because you cannot draw a line from one to the other and call it growth. A combined active-user count over a product family will be larger than a weekly count for one product in that family, by construction, before any real adoption is added on top. The honest read is that Codex-plus-ChatGPT-Work active users are now around 8 million; the honest read is not that Codex weekly users tripled. Anyone printing “Codex tripled in six weeks” is stacking two numbers that were never measured the same way.
We keep the weekly-Codex series — the apples-to-apples one — on the stats reference, and we are leaving the 8 million figure off that specific chart on purpose, because dropping a combined-product active count into a weekly-Codex column would be exactly the mistake this section is warning about.
What it does resolve
For the last week, three outlets put Codex somewhere between six and eight million within the same news cycle, with no OpenAI post to anchor any of it. Our five-hour-cap piece refused to print a figure on that basis, calling a number three outlets disagree about “a rumor with a decimal point.”
Sottiaux’s post is the anchor that was missing. It does not retroactively confirm the six- and seven-million way-points anyone cited earlier — those still trace to secondhand reporting — but it does give the current figure a named source and an exact scope. That is enough to state 8 million as reported, with the denominator spelled out, rather than as rumor.
The limits reset, and the GPT-5.6 Sol pressure behind it
The same post confirms two operational facts we had been tracking from user reports rather than from OpenAI directly. Usage limits are being reset again “for all,” and the rolling five-hour rate limit “continues to not” be enforced — the state our limits reference has been flagging since around July 12. Both are framed as making room for people to push GPT-5.6 Sol, the flagship model in the GPT-5.6 family that launched July 9.
This lines up with the standing pledge Sam Altman made back at the three-million mark: reset usage limits at every additional million weekly users, up to ten million. A reset arriving alongside this milestone is consistent with that pledge. What we still cannot do is map the reset cleanly onto a weekly-Codex milestone, because the figure driving the announcement is the combined active-user count, not the weekly one the pledge was written against.
The caveat that stays attached
Every user number Codex has ever posted traces back to OpenAI itself, and this one traces back one step further — to an engineer’s X post, not to a published announcement page with a methodology behind it. There is no independent measurement of Codex or ChatGPT Work active users, weekly or otherwise. Treat 8 million as directionally credible, precisely unverifiable, and — this time especially — scoped to a combined product family rather than to Codex alone.
We have logged the figure with its scope on the stats page and reflected the limits state on the limits reference. If OpenAI publishes an official post with a weekly-Codex figure we can compare against the June 2 number, we will update this story and the chart the same day.