Anthropic is ending free access to Claude Fable 5 on July 20, moving its most powerful model to usage-based pricing that charges $10 per million input tokens and $50 per million output tokens. This marks a significant shift in how the company structures AI subscriptions, separating everyday models from its newest flagship system. Standard Claude subscriptions will continue to include Opus 4.8, Sonnet 5, and Haiku, but users who want Fable 5 will need to enable pay-as-you-go credits in their Claude settings.

Why Is Anthropic Making This Change?

Anthropic has been offering free access to Fable 5 since July 1, allowing Pro, Max, Team, and premium Enterprise subscribers to use up to 50% of their normal weekly usage limits at no extra cost. That temporary arrangement was intended to "give customers a chance to experience Fable 5 while we continue expanding capacity," according to the company. The repeated deadline extensions suggest Anthropic has been balancing customer adoption with the realities of compute capacity constraints.

Fable 5 is not just another model update. It is Anthropic's first Mythos-class model, designed for long-running coding sessions, AI agents, and workloads that can use its one-million-token context window, which allows it to process roughly 100,000 words at once. Those capabilities also make it significantly more expensive to operate than earlier Claude models. Anthropic noted that "Fable 5 is designed for long-running agentic workflows and sustained reasoning over very large contexts".

Anthropic

Across the AI industry, frontier models are becoming increasingly expensive to train and operate. Rather than raising subscription prices across the board, companies are beginning to separate everyday models from their newest and most compute-intensive systems. Anthropic CEO Dario Amodei has argued that frontier AI models require "enormous amounts of compute" and that expanding capacity remains one of the industry's biggest challenges.

Who Will Be Most Affected by This Pricing Change?

The people most affected are developers and researchers who have made Fable 5 their default model for AI coding agents, large research projects, or other workloads that generate millions of tokens. For those users, costs can increase quickly once usage-based billing begins. However, features such as prompt caching help reduce repeated processing costs by avoiding redundant token consumption.

Everyday conversations, writing tasks, and coding assistance remain well covered by Opus 4.8 and Sonnet 5, both of which stay included in paid plans. This means casual users and those working on standard projects will not see any change to their subscription experience.

Steps to Prepare for the Fable 5 Pricing Change

• Check Your Usage Dashboard: If you have been relying on Fable 5 over the past few weeks, review your usage dashboard before the free allocation ends on July 19 at 11:59 p.m. Pacific to understand your typical token consumption patterns.
• Enable Pay-as-You-Go Billing: Users who want to continue using Fable 5 after July 20 will need to enable usage-based billing through Claude's web settings to avoid service interruption.
• Evaluate Alternative Models: Consider whether Opus 4.8 or Sonnet 5 can handle your workload, since these models remain included in standard subscriptions and may be sufficient for many tasks.

Anthropic has indicated that its long-term plan is to fold Fable 5 back into normal subscriptions once infrastructure catches up, although the company has not specified when that will happen. The repeated deadline extensions and capacity constraints suggest this transition may take several months.

This pricing model reflects a broader trend in the AI industry. Subscriptions increasingly buy access to capable AI, but the very best models are starting to look more like premium services billed by usage. That may become the new norm as AI models continue to grow larger and more expensive to run. For Anthropic, the move allows the company to manage demand while it scales its infrastructure to support Fable 5's substantial computational requirements.