anthropic.com
After three deadline extensions spread across five weeks, Anthropic has settled on a permanent access structure for Claude Fable 5 — and the resolution draws a sharp line through its paid subscriber base. Starting July 20, Max and Team Premium subscribers will receive Fable 5 as a permanent subscription feature. Pro and Team Standard subscribers will not, per Anthropic’s July 17 announcement.
The split comes with an important asterisk for Max and Team Premium users: Fable 5 is included at 50% of their standard weekly usage limits, and those standard limits will themselves shrink on the same day when the 50% boost to Claude Code weekly rate limits ends, according to the same announcement. For Pro and Team Standard subscribers, the company will issue a one-time $100 usage credit, after which continued Fable 5 access requires paying API rates of $10 per million input tokens and $50 per million output tokens — the steepest pricing for any Claude model in general availability.
Three Extensions, One Resolution
Fable 5’s path to a permanent subscription structure has been unusually turbulent. Anthropic originally planned to end free subscription inclusion on July 7, then extended the window to July 12 following user backlash, then extended again to July 19 as competitive and capacity considerations continued to shift. Each extension was announced via Anthropic’s official @claudeai account on X.
In its July 17 announcement, Anthropic acknowledged the disruption directly, noting that demand for Fable had been challenging to predict, which was why the company rolled out access to subscription plans in stages, extending access several times as it secured additional capacity.
The capacity explanation is technically substantive, not just messaging. Fable 5 belongs to Anthropic’s Mythos class — a model tier introduced above the Opus class in June 2026 that uses adaptive reasoning architecture requiring substantially more compute per token than Opus 4.8. Serving Fable 5 to a large subscriber population at full limits demands provisioning that Anthropic has been building incrementally since the model’s June 9 launch.
What Fable 5 Is — and Why It Is Expensive to Serve
Claude Fable 5 is the same underlying model as Claude Mythos 5, the version available only to a small group of vetted partners through Project Glasswing. What distinguishes Fable 5 from Mythos 5 is a layer of safety classifiers that automatically detect queries touching cybersecurity exploitation, biology and chemistry uplift, and suspected large-scale model distillation attempts. When a classifier fires, the response is handled by Claude Opus 4.8 instead of Fable 5. Anthropic reports this fallback triggers in fewer than 5% of sessions.
The architecture produces a model with a 1-million-token context window and a maximum output of 128,000 tokens per response, per Anthropic’s model specifications. Those parameters are designed specifically for tasks that require holding an entire codebase, contract corpus, or research archive in a single session and producing extended structured output — exactly the kind of autonomous, multi-hour work that makes the model token-intensive by design. Anthropic also applies a mandatory 30-day data retention policy to all Mythos-class model traffic, including Fable 5, a requirement that has no equivalent on Opus-class models.
On software engineering benchmarks, Fable 5 posted 80.3% on SWE-Bench Pro according to Anthropic’s own evaluation using Anthropic’s scaffolding — a figure that has not been independently replicated on the same benchmark by a neutral evaluator as of this writing. On the Artificial Analysis Intelligence Index, an independent composite benchmark, Fable 5 scores approximately 59.9.
Why GPT-5.6 Sol Changed the Calculation
Anthropic’s original plan had been to drop Fable 5 from subscription plans entirely, making it available only through per-token API billing. The reversal followed OpenAI’s release of GPT-5.6 Sol, which reached general availability on July 9 at $5 per million input tokens and $30 per million output tokens — half the input cost and 60% of the output cost of Fable 5’s published rates.
On the same Artificial Analysis Intelligence Index where Fable 5 scores 59.9, GPT-5.6 Sol scores 58.9 — a statistical tie in general capability. The more significant difference is per-task cost: Artificial Analysis found that Sol completes equivalent tasks at approximately one-third the cost of Fable 5, because it emits fewer output tokens to reach comparable results.
The competitive math was straightforward from a subscriber-retention standpoint. A Max or Team Premium plan that does not include Anthropic’s best available model offers little differentiation against a competitor whose flagship model costs meaningfully less per task at near-equivalent general capability.
What the $100 Credit Actually Covers
For Pro and Team Standard subscribers evaluating whether the one-time credit constitutes meaningful access, the token arithmetic is unfavorable in a specific way: the credit is sized around Fable 5’s per-token rate, but Fable 5 was designed for the kind of work that consumes output tokens at scale.
At the published API rate of $50 per million output tokens, a session that produces 2 million output tokens — modest by the standard of an autonomous multi-file coding migration or a deep document analysis run — costs exactly $100. That is the entire one-time credit in a single session. The same calculation applies to any agentic workflow where Fable 5 is doing the kind of extended, iterative generation the model’s architecture was built to support.
Anthropic has not stated when the credit will be distributed or what a typical user is expected to get from it. Pro subscribers who have been using Fable 5 regularly for intensive work face a decision that is effectively binary: upgrade to Max, migrate to a lower-tier Claude model, or shift to per-token API billing after the credit is exhausted.
The Full Pricing Landscape After July 20
The restructuring makes visible a pricing ladder that spans an order of magnitude within a single vendor. From the official Anthropic pricing documentation, the four current Claude model tiers are: Haiku 4.5 at $1 per million input tokens and $5 per million output tokens; Sonnet 5 at $2 per million input tokens and $10 per million output tokens at introductory pricing through August 31, 2026; Opus 4.8 at $5 per million input tokens and $25 per million output tokens; and Fable 5 at $10 per million input tokens and $50 per million output tokens.
The Fable 5-to-Opus 4.8 output cost ratio of 2:1 is the largest within Anthropic’s model lineup. For teams whose workloads do not specifically require the 1-million-token context window or the extended output capacity that define Fable 5’s technical differentiation, Sonnet 5 during its introductory period remains the most cost-efficient redirection available.
For subscribers in the Max and Team Premium tiers, July 20 brings permanence at a reduced rate. For Pro and Team Standard subscribers, it brings a hard cutover from the subscription model that has governed Fable 5 access since the model’s launch. Whether Anthropic eventually restores Fable 5 as a standard Pro-tier benefit — the promise a Claude Code lead engineer made publicly in early July — will depend on how quickly the company can close the gap between Fable 5 demand and available infrastructure capacity.
Frequently Asked QuestionsDoes Claude Pro still include Fable 5 after July 20?
No. Starting July 20, Pro and Team Standard subscribers no longer receive Fable 5 as a bundled subscription feature. Anthropic is issuing a one-time $100 usage credit for continued access; once that credit is exhausted, further Fable 5 use is billed at API rates of $10 per million input tokens and $50 per million output tokens. Max and Team Premium subscribers retain Fable 5 access, but capped at 50% of their standard weekly usage limits, per the July 17 announcement.
How long will the $100 credit last for typical Fable 5 use?
It depends entirely on the workload, but the math is unfavorable for heavy agentic use. At $50 per million output tokens, a single session that generates 2 million output tokens — a moderate scope for an autonomous coding migration or a document-intensive research run — exhausts the entire credit. Pro subscribers using Fable 5 for intensive work should plan for the credit to last one to a few sessions at most, not weeks of regular use.
How does Claude Fable 5 compare to GPT-5.6 Sol?
On the Artificial Analysis Intelligence Index, an independent composite benchmark, Fable 5 scores approximately 59.9 and GPT-5.6 Sol scores approximately 58.9 as of July 18, 2026 — effectively a tie on general capability. Sol costs $5 per million input tokens and $30 per million output tokens, compared with $10 and $50 for Fable 5. Artificial Analysis found Sol completes equivalent tasks at roughly one-third Fable 5’s per-task cost because it emits fewer output tokens. Fable 5 holds a narrower lead on repository-level software engineering tasks; Sol leads on terminal-based and agentic coding agent benchmarks.
Should Max subscribers be satisfied with 50% of their usual limits?
It depends on usage pattern. The 50% cap is measured against the standard weekly limit, which itself shrinks on July 20 as the 50% boost to Claude Code weekly rate limits ends on the same day, per Anthropic’s announcement. Users who need occasional or moderate frontier-class reasoning get a permanent, predictable entitlement without per-token billing. Users who relied on higher weekly volumes during the promotional period should audit their typical session count against the reduced baseline to determine whether the Max tier continues to meet their needs or whether they would be better served by a hybrid approach using Sonnet 5 for lighter tasks.