News · 2026-09-16

Anthropic ends Claude Code’s temporary 50% weekly-limit boost

Anthropic ended a temporary 50 percent weekly-limit boost for Claude Code at the end of September 13 and replaced it with a permanent 25 percent increase over the old baseline. The documented policy is narrower than “Claude subscriptions were cut,” but it leaves affected heavy Claude Code users with 16.7 percent less weekly headroom than they had during the four-month promotion.

Key facts

• The promotion ran May 13 through September 13, 2026, ending at 11:59 p.m. Pacific Time.
• From September 14, eligible users receive 1.25 times the former baseline instead of the promotional 1.5 times.
• The five-hour usage limit did not change.
• The primary source is Anthropic’s Claude Code weekly-limit announcement.

The arithmetic is important because the language can sound contradictory. Anthropic has made part of an increase permanent, so the new weekly allowance is better than the pre-May baseline. But it is lower than the operating reality users had become accustomed to: 1.25 divided by 1.5 equals 83.3 percent, a 16.7 percent reduction from the promotional level. Anthropic writes that the full boost “couldn’t last,” while saying users found the additional usage helpful.

The scope is Claude Code, not every Claude surface. Anthropic says the policy covers Pro, Max, Team and legacy seat-based Enterprise plans for Claude Code in the CLI, IDE extensions, desktop and web. Free plans and consumption-based Enterprise seats are excluded. Other Claude products, including the general web, desktop, mobile and Cowork experiences, were unchanged by this promotion. The pricing page supplies plan context, but it does not turn the policy into a universal subscription cut.

Why some builders experience this as a cliff is a product-design issue, not evidence that Anthropic changed more than it documented. Claude Code maintains conversation and project context on each turn; long sessions, large repositories and repeated tool work consume more capacity than a count of visible prompts suggests. Anthropic’s usage documentation also says Opus consumes meaningfully more quota than Sonnet. A weekly bucket and a rolling five-hour bucket are different gates: reaching a new five-hour window does not necessarily restore a weekly allowance.

The September timing also intersects with Fable 5.1. Anthropic’s release notes date it to September 1 and say it defaults to High effort in Claude Code. The Fable plan rules say premium plans can use Fable within the same regular weekly limit, where it consumes allowance faster; it is not an extra 50 percent pool. On Pro and standard Team plans, it uses purchased credits. That is a plausible documented explanation for rapid meter consumption, though it cannot diagnose an individual complaint.

The rollout has a communication wrinkle. The policy is in a current official support article, but it was not listed as a September 14 item in the release-note stream. Users could reasonably have noticed their meters first. They cannot accurately say Anthropic offered no notice, because the support article states the end date and replacement. The dossier’s same-day Reddit index records several dissatisfied thread titles, which establishes attention and frustration, not universal usability or a measured account-level reduction.

There is no downloadable model involved, no independent benchmark, and no claim of technical model quality. This is an economic and workflow story. The requested tool is becoming production infrastructure for some teams, and quota policy is therefore part of product capability. The practical response is to watch the weekly meter separately from the short-term window, choose models and effort levels deliberately, cap autonomous loops, and budget paid credits for expensive models.

The strongest counterargument is that a permanent 25 percent improvement over the old baseline is still a gain. That is true. The equally true operational fact is that capacity planning based on the temporary 50 percent boost no longer holds. Anthropic has published relative policy levels rather than a universal token or message cap, so the honest conclusion ends there: the policy change is real and precisely quantified; the absolute impact depends on each user’s model, context and workflow.

Primary source, verified: read the paper →

Ground Truth. (2026, September 16). Anthropic ends Claude Code’s temporary 50% weekly-limit boost. Ground Truth. https://groundtruth.day/news/anthropic-claude-code-weekly-limit-promotion-end.html

@misc{groundtruth:anthropic-claude-code-weekly-limit-promotion-end,
  title  = {Anthropic ends Claude Code’s temporary 50% weekly-limit boost},
  author = {{Ground Truth}},
  year   = {2026},
  month  = {sep},
  url    = {https://groundtruth.day/news/anthropic-claude-code-weekly-limit-promotion-end.html}
}

Topics: anthropic · developer-tools · usage-limits · ai-economics