Anthropic Trims Claude Code Limits by 17% in Practice

Three numbers explain Anthropic's latest Claude Code announcement: 100, 150 and 125. Call your baseline weekly limit 100 units. A temporary boost has had most users running at 150 for a while. On September 14, 2026 that boost expires and a permanent 25 percent increase kicks in, leaving you at 125. Anthropic calls it a raise; your terminal will experience it as a 17 percent cut.

The arithmetic the headline leaves out

The permanent increase is real, and it applies broadly: Pro, Max, Team and seat-based Enterprise plans all get 25 percent more than their original baseline. But the comparison that matters to a working team is not against the baseline; it is against last week. Measured that way, capacity drops from 150 to 125, and 25 divided by 150 is the 17 percent that BleepingComputer put in its headline.

Put differently: a team that spreads its Claude Code usage evenly across five working days is losing a little under one day's worth of quota per week.

Why now?

Anthropic's framing is that a temporary measure is ending and a permanent improvement is starting, which is technically accurate. Asked by TechCrunch, the company added that it is working on fixing slow response times. Those two statements together sketch the real story: demand for Claude Code has been running ahead of available compute for months, and with fixed capacity, speed and quota trade off against each other. The Decoder's reading is blunter: this is a tightening.

It also fits the summer's broader pattern: coding agents have become the fastest-growing consumers of inference capacity across the industry, and vendors keep adjusting the plumbing underneath them, sometimes upward, sometimes, as here, effectively downward.

How to tell whether you will feel it

You will only notice the change if you currently use more than 83 percent of your effective weekly limit. Most teams have no idea what that number is for them, which makes the next two weeks a good time to find out.

• Check your usage dashboard now and note your weekly consumption. Above 80 percent, start the conversation about a plan upgrade or splitting work across seats.
• Route routine tasks to smaller models. Long idle sessions and sending everything to the largest model burn quota silently.
• Audit agent workflows for waste: repeated file scans and retry loops are usually the biggest hidden line item.

Claude Code has had a busy month on this site; last week we covered how a researcher bypassed Auto Mode's protections most of the time. The pattern holds for quotas as it does for security: the defaults are not a substitute for knowing how the tool actually behaves.

Sources: BleepingComputer, Notebookcheck