OpenAI Just Lifted Codex's 5-Hour Usage Cap. Here's Why That Matters.
If you've used Codex or ChatGPT Work recently, you've probably run into the wall: burn through your usage, hit a 5-hour cooldown, wait it out. OpenAI just pulled that wall down — temporarily — and paired it with efficiency upgrades to GPT-5.6 Sol that make the whole thing cheaper to run in the first place.
Here's what actually shipped.
The Three Changes
OpenAI rolled out three updates to Codex and ChatGPT Work at once:
- The 5-hour usage limit is temporarily gone for Plus, Business, and Pro plans. No more waiting out a cooldown window mid-task.
- GPT-5.6 Sol got broad efficiency improvements, meaning it now uses fewer tokens to do the same work. OpenAI hasn't published hard numbers on how much cheaper this makes a typical session, but the direction is clear: less token spend per task.
- Usage resets now happen hourly instead of on the old 5-hour cadence, giving active users a faster refresh cycle even once the temporary cap removal ends.
OpenAI also disclosed that Codex has crossed 6 million active users — a scale marker that puts real weight behind why they're loosening limits now rather than later.
Why Remove the Cap At All
Usage caps exist for one reason: cost control. Coding agents like Codex run long, expensive sessions — reading files, running tools, iterating on code — and a 5-hour limit was OpenAI's way of keeping that in check.
Pairing the cap removal with a token-efficiency upgrade to GPT-5.6 Sol isn't a coincidence. If the model itself burns fewer tokens per task, OpenAI can afford to let people run longer without the same cost blowout. It's less "we're giving away compute" and more "we made the compute cheaper, so the ceiling can go up."
That's worth noting if you've been budget-conscious about agent usage. A model that's meaningfully more token-efficient changes the math on what's worth running as an autonomous, multi-step task versus what you keep doing by hand.
What This Signals About Coding Agents
Six million active Codex users is a real number, not a beta cohort. That's a sign coding agents have moved past the "interesting experiment" phase into daily-driver territory for a meaningful chunk of developers.
It also tracks with something that keeps showing up across the industry right now: usage limits and token costs are the actual bottleneck for agentic workflows, not model capability. Models can already plan, use tools, and iterate — what's held people back is the meter running out mid-task. Every efficiency gain that pushes that limit back is a direct unlock for longer, more ambitious agent runs.
What This Means If You Use OpenClaw
OpenClaw runs on the same underlying dynamic: the more efficient the model, and the fewer artificial ceilings in the way, the more your agent can actually get done in one sitting.
When you run a tutorial on ClawWorld, your agent isn't just answering one question — it's working through multi-step tasks, calling tools, and building on what it did earlier in the session. That's exactly the kind of workload that benefits most from efficiency gains like the ones GPT-5.6 Sol just got, and from usage limits that don't cut a task off halfway through.
You don't need to track any of this yourself. OpenClaw is built to make the most of whatever headroom the underlying models give it, so your agent keeps working instead of hitting a wall.
The Bigger Picture
Coding agents are hitting a scale where usage limits and token efficiency matter as much as raw capability. OpenAI loosening Codex's cap — backed by real efficiency work under the hood — is a small update on paper, but it's a preview of where the whole category is headed: agents that can run longer, do more, and cost less to do it.
If you want to see what an agent can actually accomplish when it's not fighting the clock, that's what OpenClaw is for.