Anthropic has changed Claude Code weekly limits for paid subscribers. From 14 September 2026, the standard weekly allowance for Pro, Max, Team and seat-based Enterprise plans is permanently 25% higher than the baseline that applied before Anthropic’s temporary capacity promotion.
That headline needs context. During the promotion, eligible users had a 50% boost over the old baseline. Moving from 150% of baseline to 125% is therefore an effective reduction of about 16.7% compared with the allowance available immediately before the change. Users keep more capacity than they originally had, but less than they had during the temporary boost.
Background: why Claude Code has usage limits
Claude Code is Anthropic’s agentic coding tool for terminals and development environments. It can inspect repositories, edit multiple files, run commands and work through lengthy software tasks. Those workflows can consume substantially more model capacity than a short chatbot exchange, especially when a session uses large codebases, subagents, extended reasoning or repeated test cycles.
Anthropic uses several controls to manage that demand. Subscription users can encounter shorter-window and weekly allowances, while API customers have separate rate and spend limits. Anthropic’s platform documentation explains that API limits are applied to manage capacity and misuse, and that published maximums are not guaranteed minimum throughput. That API system should not be confused with the subscription-plan weekly change covered here.
What changed on 14 September 2026
Anthropic said the permanent standard weekly limit for Claude Code would rise by 25% for Pro, Max, Team and seat-based Enterprise plans. The change replaced a temporary promotion that had raised weekly limits by 50%.
A simple baseline makes the difference clearer:
- Original standard allowance: 100 units
- Temporary promotional allowance: 150 units
- New permanent allowance: 125 units
So the permanent level is genuinely 25% better than the original standard. However, it is 25 units lower than the temporary 150-unit level. Dividing that difference by 150 produces a 16.7% effective reduction from the capacity users had just before 14 September.
These units are illustrative, not token quotas. Anthropic does not present every subscriber with one simple universal token number because practical usage varies with the selected model, conversation length, repository context and the work Claude performs.
Why the Claude Code weekly limits change matters
Heavy users may notice the difference first
Developers who regularly reached the temporary weekly ceiling have less headroom now. Long agentic sessions, parallel work across repositories and high-context debugging are the most likely to make the reduction visible. Occasional users may notice little or no change.
The permanent baseline is still higher
The change is not simply a return to the previous limit. Keeping a 25% uplift provides more sustained capacity than the pre-promotion standard. Anthropic had already doubled Claude Code’s five-hour rate limits for the same plan groups in May 2026 and removed peak-hour reductions for Pro and Max accounts, citing expanded compute capacity.
Capacity planning is becoming part of AI development
AI coding tools are increasingly used for multi-step work rather than isolated code completion. Teams now need to plan model capacity much as they plan CI minutes, cloud budgets and API quotas. A workflow that depends on an assistant being continuously available can stall when limits are reached, even if the underlying code and infrastructure are ready.
Practical steps for developers and teams
- Check usage inside Claude Code. The
/usagecommand shows plan-limit information and can help identify which activities are consuming capacity. - Start fresh sessions deliberately. Very long conversations carry more context on each turn. Preserve essential decisions in project documentation, then begin a focused session when old context is no longer useful.
- Use the right model for the task. Reserve the most compute-intensive option for architecture, difficult debugging and complex agentic work. Routine edits may not need it.
- Keep prompts and repository scope focused. Ask the tool to inspect relevant directories rather than an entire monorepo when possible, and provide clear acceptance criteria before it starts.
- Monitor team patterns. Team and Enterprise administrators should look for repeated workflows, oversized context, unnecessary parallel agents or jobs that could run through a direct API with controlled budgets.
- Maintain a fallback. Critical releases should not depend on one AI service or an unverified amount of weekly capacity. Keep conventional development and review processes workable.
Risks, limitations and common misunderstandings
The biggest risk is reading “25% increase” without its baseline. It describes the new permanent allowance relative to the old standard, not relative to the temporary 50% promotion. Both statements—higher than the original limit and lower than the immediately preceding limit—can be true.
Usage also does not map neatly to a fixed number of prompts. A small request with a large repository and long session history may cost more capacity than several concise requests. Model choice, tool calls and generated output all affect how quickly an allowance is consumed.
Finally, subscription limits and Claude API rate limits are separate systems. Moving work to the API can provide different controls and billing, but it is not a free bypass: organisations still face tier, rate and spend constraints.
What to watch next
Users should watch Anthropic’s official documentation and in-product usage display for plan-specific clarification. The key question is whether growing compute supply leads to further permanent increases, or whether demand from larger models and more autonomous coding workflows absorbs that capacity.
It is also worth monitoring how competing AI coding services explain quotas. Clear, comparable capacity metrics could become an important purchasing factor alongside model quality, privacy, price and integration support.
Conclusion
The September change leaves Claude Code weekly limits in an in-between position: 25% above the old permanent baseline, yet about 16.7% below the temporary promotional level. For most developers, the sensible response is not to chase an assumed prompt count. Use the built-in usage view, control context growth and reserve expensive workflows for tasks where the coding agent delivers the most value.