Anthropic extends 50% higher Claude Code weekly limits to August 19

• Claude Code users on Pro, Max, Team, and seat-based Enterprise keep 50% higher weekly limits through August 19, according to ClaudeDevs' announcement.
• The extension delays a nasty July 20 cliff because bilawalsidhu's math put Fable at roughly a third of the May-to-July experience once the 1.5x weekly promo expired.
• The 5-hour cap remains separate, and LLMJunky's one-prompt run burned 100% of a session while using only 20% of the weekly limit.
• Demand was high enough that trq212's follow-up said full-time subscription access took “a lot of work” just to avoid a Max-plan gap.
• Creators are already building limit telemetry, including om_patel5's iPhone widgets for Claude Code, Codex, and Cursor.

The updated Claude Help Center page says the 50% weekly boost applies automatically and can be checked with /usage. Anthropic's May compute post said a SpaceX deal helped double Claude Code's five-hour limits in May, while the current Claude usage docs separate that session wall from the weekly “conversation budget.” On Hacker News, one commenter said Fable on a $20 plan would likely wreck quota, and another asked Anthropic to extend the weekly boost to stay competitive. Meanwhile, om_patel5 found the creator-tool answer to invisible agent quotas: lock-screen widgets.

ClaudeDevs said the weekly limit promo now runs through August 19 for Pro, Max, Team, and seat-based Enterprise users.

The updated Claude Help Center page makes the boundaries sharper:

• Eligible: Pro, Max, Team, and legacy seat-based Enterprise users.
• Excluded: Free plans and consumption-based Enterprise seats.
• Surfaces: Claude Code in the CLI, IDE extensions, desktop, and web.
• Scope: weekly Claude Code limits only, not the 5-hour usage limit.
• Check: /usage in the CLI.

A July 13 Help Net Security writeup captured the prior deadline as July 19 at 11:59 PM PT. bilawalsidhu caught the new page shortly after it landed, with “Updated in the last hour” visible in his screenshot.

The plan labels got weird because Fable access was being discussed as 50% of limits. bilawalsidhu translated that bluntly: the 20x plan becomes 10x, and the 5x plan becomes 2.5x.

In bilawalsidhu's follow-up, the second denominator was the expiring 1.5x weekly boost: “50% of limits” would be one-third of what subscribers had been using since May.

• 20x Max, with Fable at 50%: 10x.
• 5x Max, with Fable at 50%: 2.5x.
• 20x Max, after losing the 1.5x boost: 6.67x versus the recent promo ceiling.
• 5x Max, after losing the 1.5x boost: 1.67x versus the recent promo ceiling.

The August 19 extension delays the second haircut, which is why “someone found Anthropic some compute” became the cleanest read in bilawalsidhu's update.

In trq212's post, the internal story was capacity triage: people worked “literally around the clock,” and it was not clear Anthropic would finish in time.

The follow-up from trq212 says the goal had been full-time subscription access, but demand was high and the work was needed to avoid a Max-plan gap.

The backlash did not produce full restoration. alliekmiller's read framed the result as compute-constrained: Fable stayed on the highest tiers with lower limits even after the subscription backlash and the opening it handed OpenAI Codex.

The weekly promo did not change the session cap. The help page says the 5-hour usage limit is unaffected, and Claude's usage docs say usage varies by conversation length, model, features, and effort level.

That split is exactly what LLMJunky's complaint hit: 100% of a five-hour window and 20% of weekly usage on one unfinished prompt.

The same shape appears in thekitze's usage screenshot: current session exhausted, weekly all-model usage at 20%, and Fable weekly usage at 33%.

On Hacker News, one $20-plan user said a few Opus requests used the full five-hour limit, before adding that they did not want to know what Fable would do.

The painful case, according to LLMJunky, is leaving a goal running and getting locked out for six days.

LLMJunky's Cursor comparison tied the same problem to long-running loops: the cutoff feels worse when it lasts for days.

That is the product shape behind the widget boom. Coding agents run while the user is away, then the quota failure becomes a stalled task instead of a visible error.

om_patel5 posted iPhone widgets that turn Claude Code limits into a home-screen and lock-screen UI.

The demo tracks:

• Claude Code session windows.
• Weekly Claude Code limits.
• Fable limits.
• Codex, Claude Code, and Cursor side by side.
• Lock-screen progress bars and circular rings.
• Notifications when usage depletes or fully resets.
• Local credential storage in the iPhone keychain, with no server tracking claimed in om_patel5's demo.

Tiny limit widgets are the most honest artifact of the week: AI agents have become long-running enough that quota became ambient infrastructure.