Anthropic Retracts and Extends Fable 5 in Subscriptions Until July 12 After Subscriber Reaction
Hours before the midnight cutoff on the Pacific Coast, Anthropic extended included access for five days and limited usage to 50% of the weekly quota. The more expensive model in the catalog continues to face capacity issues.
Anthropic partially retracted its decision to remove Claude Fable 5 from the Pro, Max, Team plans and part of the Enterprise contracts starting at midnight on July 8 in the Pacific time zone. Just hours before the cutoff, the company announced an extension of included access until July 12, with a limit of 50% of the weekly usage quota per subscriber. After that, anyone wishing to continue will need to purchase usage credits at the official rate of $10 per million input tokens and $50 per million output tokens, double that of Opus 4.8 and the highest token price that Anthropic has ever published for a model in general availability.
The sequence of events, summarized over three days, helps explain why the decision sounded misaligned. On July 6, engineers at Anthropic circulated a note reminding that July 7 would be the last day of included access. On July 7, amidst an avalanche of reactions on X and the official Claude forum, the company published a blog post titled "Redeploying Claude Fable 5" announcing the extension until July 12 and the 50% rule. A lead engineer of Claude Code publicly stated that the company intends to restore Fable 5 as a standard part of subscriptions as soon as capacity allows, without providing a timeline.
What the Numbers Say
Fable 5, at $10 per million input tokens, costs twice that of Opus 4.8. In comparison with open-source Chinese models that have entered the American market strongly in recent weeks, the base GLM 5.2 is sold at $0.95 per million inputs, or approximately 9% of the price of Fable 5. This calculation helps explain why Anthropic attempted to remove the model from the bundle: maintaining Fable 5 within a $200 monthly subscription for the Max plan requires subsidizing a volume that does not exist among average users but explodes among heavy users in code and research.
The second data point, now from the revenue side, shows that the problem is not a lack of demand. In May, Anthropic stated that it was on track for $47 billion in recurring revenue by 2026 and for profits in 2029, a year ahead of OpenAI. Fable 5 and Mythos 5 are the premium base of this figure. What the company is discovering, in practice, is that compute capacity remains extremely expensive and the bottleneck is now physical, not a matter of commercial will.
Where the Decision Hits: United States, United Kingdom, and Germany
In the United States, engineering teams at companies that have standardized code review flows based on Claude Code need to decide by July 12 whether to accept the variable cost, reroute part of the traffic to cheaper models, or pause internal pipelines. In the United Kingdom, magic circle law firms and integrators like Capgemini UK, who structured Enterprise Claude pilots for financial clients, now need to recalibrate commercial proposals that assumed fixed costs. In Germany, where banks and industries entered Anthropic Enterprise through a partnership with SAP for use in Joule and Business Suite integrations, the sourcing area has just gained ammunition to renegotiate contracts that have not yet been finalized.
The message that CIOs will take away from the week is that the flat subscription economy for frontier models is not sustainable at the current pace. This point applies to those operating with OpenAI, Anthropic, and Google. Each of these companies is rewriting contract terms in 2026 based on the same stressed capacity equation.
The Steelman and What Remains Standing
Those defending Anthropic have a solid argument. Charging for consumption in frontier models is how the entire sector already operates at the API layer, and Fable 5 was never marketed as an unlimited flat rate service. The five-day extension, the company signals, is a commercial courtesy, not an admission of error. The counterargument, also valid, is that the initial announcement came too late for engineering teams to adjust budgets and pipelines within the current sprint cycle. The decision to communicate the change only 48 hours before the cutoff is the source of the reaction, more than the price itself.
What remains standing at the end of the week is an uncomfortable realization: the market's most premium model has once again become a rationed good, and the prioritization queue within each company is the next debate.