OpenAI has changed Codex and ChatGPT Work’s usage limits three times in a month. The five-hour rate cap was removed in late July, paid ChatGPT Work and Codex limits were reset for all users on 8 August, and the five-hour cap then returned for Plus accounts on 26 August. Each change has been announced with little notice and reversed or adjusted within weeks, and OpenAI’s own developer community forum has run a steady thread of users asking what quota period they’re actually on.
This isn’t really a Codex story — it’s a capacity story. Every AI coding vendor is managing the same underlying problem: demand for agentic coding sessions is growing faster than the compute committed to serve it, and usage limits are the lever vendors pull first when that gap opens up. Anthropic has run its own version of this with Claude Code’s weekly limits over the summer, extending and rolling back caps in response to demand spikes. The difference worth noting is frequency — three changes to the same product’s limits in a month is a faster cycle than most teams plan around.
Why this matters if you’re not the one hitting the limit
If your organisation is standardising on an AI coding tool, or a development partner is building your product with one, the vendor’s usage-limit stability is part of the actual cost and delivery-risk picture — not a footnote. A team that hits an unannounced rate cap mid-sprint loses velocity in a way that’s hard to plan around, and a tool whose limits change three times in a month is telling you something about how tightly its provider’s capacity is stretched right now.
So what
We treat AI coding tool selection as a scoping decision, not a developer preference — which vendor, which plan, and how exposed a build is to that vendor’s usage-limit volatility all factor into how we plan delivery timelines. See our AI-assisted development approach, or get in touch if you want that risk built into your project plan from day one.