Anthropic has officially extended promotional free access to its frontier reasoning model, Claude Fable 5, until July 12, 2026, at 11:59 PM PT.

The extension offers a brief reprieve to paid subscribers on Pro, Max, Team, and eligible Enterprise tiers following initial pushback over a tight usage window.

1. Terms of the Extension

The promotional access operates under specific caps to help Anthropic manage capacity:

  • The 50% Cap: Eligible subscribers can use Claude Fable 5 for up to 50% of their existing weekly subscription limit at no additional charge.
  • Consumption Speed: While it draws directly from your regular weekly usage bucket, Fable 5 consumes that allocation at a significantly faster rate than models like Sonnet or Opus due to its heavy backend processing.
  • Exhaustion Protocol: Once you cross the 50% threshold within a week, you must either purchase separate usage credits to keep using Fable 5 or switch back to another Claude model to remain within your normal plan constraints.

2. Transition to Pay-Per-Token Billing

The extension acts as a final testing sandbox before Anthropic shifts the model to a permanent premium pricing tier. After the July 12 cutoff, Fable 5 will be completely decoupled from standard weekly subscription limits. Continued access on the web, desktop, and mobile apps will require prepaid usage credits billed at:

  • Input Tokens: $10 per million
  • Output Tokens: $50 per million (exactly double the cost of Claude Opus 4.8)

3. Navigating a Turbulent Rollout

The extended free window follows a highly unusual launch sequence for the model:

Plaintext

[ CLAUDE FABLE 5 TIMELINE ]

├── June 9, 2026  ──► Initial global launch as Anthropic's most powerful model
├── June 12, 2026 ──► Pulled globally following US Dept of Commerce export control concerns
├── July 1, 2026  ──► Redeployed to the public with modified safety classifiers
└── July 7, 2026  ──► Promotional access extended to July 12 after user capacity backlash

4. Why Teams Are Rushing to Test It

Fable 5 represents Anthropic’s “Mythos-class” architecture designed for long-horizon autonomous engineering rather than quick chat replies. It features a 1-million token context window and a 128k output limit, powered by an “always-on” adaptive thinking routine.

Early enterprise testers, including Stripe, reported that the model can handle multi-day agentic projects, such as executing full-codebase migrations and writing its own unit tests via vision verification loops—tasks that typically take human development teams weeks to map manually. For developers and enterprises, this extended window serves as a zero-marginal-cost opportunity to stress-test these long-context workflows and calculate post-July 12 token budgets before the paywall locks into place.

Get the day’s top stories in your inbox

One concise email. No spam, unsubscribe anytime.