Anthropic has extended the 50% increase to Claude Code weekly limits again, this time through August 31, 2026. The announcement from @ClaudeDevs adds something new: “We hope to make this a permanent change to our plans.” That single sentence turns a promo into a policy question. This is the fourth deadline the boost has survived, and each extension has landed right as the previous one was about to expire.
We covered the original promotion when it launched in our breakdown of the 50% weekly limit increase. This post picks up where that one left off: what the boost covers, the full extension timeline, how to read the capacity warning Anthropic attached, and what to do if you burn through your quota anyway. One preview of the practical advice: a chunk of most people’s Claude Code usage goes to API trial-and-error the agent didn’t need to do. Verifying endpoints in Apidog first, then handing the agent a confirmed spec, keeps those tokens for work that needs a model.
What exactly is boosted
The promotion raises weekly Claude Code limits by 50% for these plans:
- Pro
- Max (both the 5x and 20x tiers)
- Team
- Enterprise, for seat-based users only
Free plans and consumption-based Enterprise seats are excluded. Nothing is required on your end; the higher ceiling applies automatically to eligible accounts, and Anthropic’s usage panel reflects it.
Scope matters here, and it’s narrower than people assume:
- The boost applies to Claude Code only: the CLI, IDE extensions, the desktop app, and Claude Code on the web.
- It raises the weekly limit. The 5-hour rolling session limits are unchanged, so you can still hit a short-term wall in a heavy afternoon even with weekly headroom to spare.
- Regular Claude chat usage isn’t part of the deal.
If you’re unsure how the weekly and session ceilings interact on your tier, our guide to Claude Pro limits walks through the mechanics; the same structure applies up the plan ladder.
The extension timeline
The boost has now been extended three times since the original window closed:
| Date announced | New end date | Notes |
|---|---|---|
| July 2026 (original promo) | ~July 13, 2026 | Launch of the 50% weekly boost |
| July 13, 2026 | July 19, 2026, 11:59 PM PT | First extension, reported by Help Net Security |
| July 19, 2026 | August 19, 2026 | Announced alongside Fable 5 staying on Max and Team Premium |
| August 19, 2026 | August 31, 2026 | Current extension, with the “hope to make this permanent” language |
Read the pattern. Each extension arrived on the day the previous deadline hit. Anthropic isn’t planning these as one long promo; it’s re-deciding every few weeks based on how much compute it has. The August 31 announcement is the first to say the quiet part: the company wants the higher limits to be the new standard and keeps stopping short of committing.
That’s a meaningful shift from mid-2025, when Anthropic was moving in the opposite direction and introducing weekly caps to rein in power users. A year later it’s competing to give usage away. Subscription plans like Claude Max are now the battleground where coding-agent vendors fight for developer loyalty, and usage ceilings are the most visible term in the contract.
How to read the capacity caveat
The full caveat in the announcement: “strong demand for our models means that capacity may be tight over the coming weeks.”
Two honest readings:
The straightforward one. Anthropic sells more inference than it can comfortably serve. Making a 50% limit increase permanent means permanently reserving compute for subscription users, and the company won’t sign that check until new capacity comes online. Extending in two-week chunks lets it walk the promise back without breaking one.
The competitive one. The Hacker News thread on the extension (255 points) is full of developers doing exactly what Anthropic fears: comparing weekly quotas across Claude Code, Codex, Kimi, and DeepSeek, and switching when the math favors someone else. Several commenters note the switching cost between coding agents is near zero now. In that market, letting limits snap back 33% overnight is a churn event. The extensions look less like generosity and more like a company that can’t afford the alternative.
Both readings can be true at once. The practical takeaway is the same either way: plan as if the boost is temporary, and treat anything above the standard limit as a bonus. If your workflow only functions at 150% of standard quota, you have a workflow problem, not a plan problem.
What to do if you still hit limits
Even with the boost, heavy users hit walls. Here’s what works, in rough order of effort.
Track where the tokens go. You can’t fix what you don’t measure. Community tools surface burn rate per session and per project; we compared the options in our Claude Code usage monitor guide. Most people who instrument their usage find a small number of long, flailing sessions eating most of the week.
Batch work around the reset. Weekly limits reset on a rolling schedule, and knowing yours changes how you plan. Push exploratory, token-hungry work to the start of a fresh window; keep the tail of the window for small, surgical tasks. Our post on the Claude Code token window reset covers how to find your reset time and schedule around it.
Stop paying agent prices for deterministic work. This is the biggest lever, and the least discussed. Watch what Claude Code does when it integrates against an API it hasn’t seen: it guesses an endpoint shape, writes a curl call, misreads the error, rewrites the call, and loops. Every iteration is model inference spent discovering something a spec already knew. That’s not agent work; that’s a lookup.
The fix is to move API verification out of the agent loop:
- Design and debug the endpoint in Apidog first. Send real requests, confirm auth, capture actual response shapes.
- Mock what doesn’t exist yet. Apidog’s mock server gives the agent a stable target, so it codes against realistic responses instead of hallucinating them.
- Connect the spec directly. The Apidog MCP Server exposes your API documentation to Claude Code, so the agent reads exact schemas instead of guessing field names. Zero-token certainty beats expensive trial and error.
- Push regression testing to the CLI. Apidog CLI runs your test scenarios in CI or a terminal for free, so you’re not spending weekly quota re-verifying endpoints an agent already integrated.
The pattern generalizes: let the model do the reasoning, and let tools do the checking. Download Apidog and try the split on your next integration; the difference shows up in your usage panel within a week.
Know your fallback. If you exhaust the weekly limit mid-sprint, the escape hatch is pay-as-you-go API access, which bills per token with no weekly cap. It has its own throttles, though; see our guide to Claude API rate limits before you rely on it.
The competitive context, honestly
The HN thread’s dominant theme wasn’t gratitude for the extension. It was token efficiency. Developers argue Claude’s verbose style burns more tokens per task than Codex’s terser output, which makes raw limit numbers hard to compare across vendors. Others report moving parts of their workload to Kimi K3, DeepSeek, or GLM, where per-token pricing is lower, and keeping Claude for the hardest reasoning work.
That critique deserves a straight answer rather than a dismissal. Verbosity is partly a capability tradeoff (more explicit reasoning, more tokens) and partly waste, and users can’t tune it much. If Anthropic wants the 50% boost to read as a limit increase rather than compensation for token-hungry outputs, output efficiency has to improve alongside quota.
For working developers, the sane posture is portfolio thinking: match the model to the task, and make your tooling vendor-neutral. An API-first workflow helps here too. A verified spec in Apidog works identically whether the agent consuming it is Claude Code, Codex, or a local model; you swap the reasoning engine without redoing the verification layer.
Where this leaves you
The 50% boost is live through August 31, 2026, and Anthropic has said out loud it wants to keep it. It hasn’t committed, and the capacity caveat is real, so budget against standard limits and enjoy the headroom while it lasts. Instrument your usage, schedule heavy work after resets, and pull deterministic API verification out of the agent loop so your quota goes to problems that need a frontier model. If the boost becomes permanent, you’ll have a lean workflow and spare capacity. If it doesn’t, you’ll barely notice the snap-back.
FAQ
Which plans get the 50% Claude Code limit increase?
Pro, Max, Team, and seat-based Enterprise plans. Free plans and consumption-based Enterprise seats are excluded. The increase applies automatically; there’s nothing to enable.
Does the boost apply to Claude chat or the API?
No. It covers Claude Code only, across the CLI, IDE extensions, desktop app, and web. Regular chat usage and API rate limits are separate systems; the Claude API rate limits guide explains how the API side is throttled.
Will the higher limits become permanent?
Anthropic says it hopes so but hasn’t committed, citing tight capacity from strong demand. The promotion has already been extended three times (July 13 to July 19, then August 19, now August 31), so a further extension or a permanent change are both plausible. Plan around standard limits to be safe.
Do the 5-hour session limits change too?
No. The promotion raises weekly limits only. The 5-hour rolling limits stay the same, so short bursts of heavy usage can still throttle you even when weekly quota remains.
