The billing change Anthropic announced on June 30 arrives tonight. After midnight Pacific Time — 3:00 AM ET on July 8 — Claude Fable 5 will no longer draw from the subscription limits of any paid plan. Every Pro, Max, Team, and Enterprise subscriber who wants to keep using Fable 5 will need usage credits enabled and funded, or the model simply will not respond.

The meter start is the most significant pricing action Anthropic has taken since the model launched on June 9. Fable 5 bills through usage credits at $10 per million input tokens and $50 per million output tokens — exactly double the rate of Claude Opus 4.8 and the highest per-token price Anthropic has ever listed for a generally available model. A class action filed in federal court last month, alleging that Anthropic's Max subscription tiers deliver less usage than advertised, adds context to how subscribers are likely to receive yet another billing change on the same model.

How Fable 5 Spent Six Weeks Getting to This Billing Cliff

Fable 5 launched June 9 as the first publicly available model from Anthropic's Mythos class — the same underlying weights as the restricted Mythos 5, but equipped with classifiers that reroute sensitive cybersecurity and biology requests to Claude Opus 4.8 rather than blocking them with refusals. The model was included at no extra cost on Pro, Max, Team, and seat-based Enterprise plans through June 22, after which usage-credit billing was supposed to begin.

Three days after launch, the billing transition was overtaken by events. On the evening of June 12, Commerce Secretary Howard Lutnick's Bureau of Industry and Security issued an Is-Informed Letter to Anthropic requiring an export license before any foreign national — including Anthropic's own non-citizen employees — could access either Fable 5 or Mythos 5. The stated basis was a report from Amazon researchers who had found a technique for bypassing Fable 5's safety classifiers by framing requests as defensive code review. The model, prompted to read a codebase and identify software flaws, in one case produced code demonstrating how a discovered vulnerability could be exploited.

Anthropic suspended both models globally within about 90 minutes of receiving the letter — the first time the US government had applied export control authority to a commercial AI model rather than to the chips used to train it.

The suspension lasted 19 days. By the time Anthropic published its redeployment post on June 30 and restored global access on July 1, Fable 5's original free-inclusion window had been consumed entirely. Anthropic responded by giving subscribers a compensatory six-day window — 50 percent of weekly usage limits through July 7, at no extra cost — before tonight's billing transition. That six-day window ends at midnight PT.

The jailbreak that triggered the shutdown, it turned out, did not expose capabilities unique to Fable 5. Anthropic's own testing found that every model it evaluated — including Claude Haiku 4.5, Sonnet 4.6, Opus 4.6, Opus 4.7, Opus 4.8, GPT-5.4, GPT-5.5, and Kimi K2.7 — could produce the same vulnerability-exploitation demonstration. The Commerce Department lifted the export controls on June 30, but not before Anthropic trained and deployed a new classifier specifically targeting the Amazon-described technique, which now blocks that approach in more than 99 percent of cases. Researchers at the Commerce Department's own Center for AI Standards and Innovation independently validated those safeguards.

The trade-off is a higher rate of false positives: the new classifier flags more legitimate coding and debugging requests as potentially sensitive, rerouting them to Opus 4.8 instead. Independent testing by BridgeMind, an AI evaluation platform, published July 2, found that only 3 of 12 TypeScript debugging tasks reached Fable 5 in practice after the relaunch — the other 9 were silently rerouted to Opus 4.8, which BridgeMind's methodology scored as zero since the model under evaluation did not complete the task.

What the $10/$50 Token Rate Actually Costs in Practice

From July 8, Fable 5 access operates through Anthropic's usage-credit system — prepaid dollars drawn down at API rates as each request processes. Credits apply across Claude.ai conversations and Claude Code terminal sessions from the same pool. Opus 4.8, Sonnet, and Haiku remain inside normal subscription limits and are unaffected by the change.

The cost math for casual users is modest. Heavy daily chat use — roughly 200,000 input tokens and 50,000 output tokens per day — runs approximately $4.50 per day, or around $135 per month on top of the existing subscription fee. For developers running Fable 5 inside agentic loops, the exposure is materially different. One reviewer, testing Anthropic's own agentic workflow mode, exhausted a $100 daily credit allocation in nine minutes, according to SecurityWeek's Fable 5 developer feedback. Agentic coding sessions that run Fable 5 against a large codebase — reading files, editing code, running tests, debugging failures — can process millions of tokens per session, producing charges that reach three figures per day in intensive use.

The per-token rate makes cost-reduction levers worth understanding before the billing clock starts.

Prompt caching reduces input costs by 90 percent on any content that recurs across requests within a session. The mechanism works at the infrastructure level: when you mark a stable portion of your prompt with a cache_control breakpoint, Anthropic's servers store the processed key-value attention states for that content rather than recomputing them on every API call. Writes cost 1.25 times the standard input rate; reads cost 10 percent of the standard input rate. For Claude Code sessions where the same codebase context, system prompt, or tool definitions appear on every turn, caching is the single highest-return cost lever available. The cache has a five-minute default time-to-live, with a one-hour option available at a higher write rate. One operational requirement: variable content — timestamps, per-request IDs, the current user message — must appear after all cached content, not before. Anything after a variable element does not cache. Full details are in Anthropic's prompt caching documentation.

The Batch API applies a 50 percent discount to both input and output rates, bringing Fable 5 to $5 per million input tokens and $25 per million output — equivalent to Opus 4.8's standard rate. For pipelines that process documents, analyze codebases, or run evaluations where responses are not needed in real time, batch processing eliminates the entire Fable 5 price premium over Opus. See Anthropic's Batch API documentation for implementation details.

Model routing is the third lever. Claude Sonnet 5, which launched June 30 as Anthropic's new default for Free and Pro plans, offers near-Opus-4.8 performance on agentic tasks at $2 per million input tokens and $10 per million output tokens — introductory pricing in effect through August 31, 2026, after which standard rates of $3 and $15 per million tokens apply. Fable 5's specific advantages over Sonnet 5 are its 1-million-token context window, its 128,000-token maximum output, and its stronger performance on the longest and most complex agentic tasks. For workloads that do not require those properties — most document analysis, shorter coding tasks, conversation — Sonnet 5 delivers the bulk of Fable 5's capability at roughly one-fifth the cost. Developers using Sonnet 5 as a Fable 5 substitute should note that Sonnet 5 uses a revised tokenizer that may process the same input as 1.0 to 1.35 times as many tokens as the previous tokenizer. Measure actual consumption before committing routing decisions.

Why This Billing Change Follows a Capacity Problem Anthropic Has Not Yet Solved

Anthropic's stated reason for moving Fable 5 to usage-credit billing is demand it cannot yet serve at subscription-inclusion scale. A Claude Code lead engineer said on X: "While it will come off subscriptions after July 7th, we aim to restore Fable as a standard part of our subscriptions as soon as capacity allows, as we mentioned in our original blog post." No timeline has been given.

The capacity constraint is real and documented. In May, Anthropic announced its SpaceX Colossus 1 compute deal — securing more than 300 megawatts across 220,000 Nvidia GPUs — after users reported hitting Claude Code usage limits far faster than expected, with some reporting that a single prompt consumed 10 percent of a weekly allowance. That capacity expansion delivered visible changes: doubled Claude Code five-hour rate limits, removed peak-hour restrictions on Pro and Max accounts, and expanded API rate limits. Even so, Fable 5's demand appears to exceed what the additional capacity can support at flat subscription pricing.

The pending class action filed by Karl Kahn on June 14 in the Northern District of California is not directly about Fable 5, but its claims are directly relevant to the current moment. Kahn alleges that Anthropic marketed the Claude Max 5x and Max 20x plans — at $100 and $200 per month respectively — as providing five and twenty times the usage of Claude Pro, while the actual limits subscribers encountered were lower than that, and usage calculation was opaque enough that subscribers could not verify what they were receiving. Anthropic declined to comment on the suit. The case is pending. The lawsuit's core concern — that the value of a Claude subscription is difficult to audit from the outside — applies with equal force to the Fable 5 transition, in which "up to 50 percent of weekly usage limits" has not been translated by Anthropic into any public figure for tokens, messages, or dollar value per plan.

Export Controls Revealed a Risk That Survives the Billing Change

The most significant implication of the past six weeks is one the billing announcement does not address. The June 12 export-control directive demonstrated that any Anthropic frontier model with sufficient dual-use potential can be withdrawn from global service by administrative order, with no advance notice, no contractual remedy, and no recourse period for enterprises or developers who have built production dependencies on it. That risk did not disappear when the export controls were lifted on June 30. It is an ongoing structural condition for any organization whose workflows depend on Fable 5 or any future Mythos-class model. Mayer Brown's legal analysis of AI export controls details the regulatory basis and precedent set by this episode.

Anthropic itself has acknowledged the system needs to change. As part of the redeployment agreement, Anthropic, Amazon, Microsoft, Google, and other Project Glasswing partners began developing a shared industry framework for assessing and classifying potential jailbreaks — analogous to the Common Vulnerability Scoring System used in software security — so that AI developers can triage reported bypasses consistently, launch capable models with greater safety, and communicate risk levels to government partners without triggering the kind of surprise shutdown that took Fable 5 offline in June.

For enterprises evaluating whether to build deep dependencies on Fable 5, that framework does not yet exist in finalized form. A dependency audit — identifying every production touchpoint where Fable 5 is called directly, through Claude Code, or through the API — is the practical risk-management step available now, separate from any immediate action required on usage credits.

How to Enable Credits Before Tonight's Deadline

Usage credits are enabled at claude.ai → Settings → Usage. Subscribers who purchased their plan through the iOS or Android mobile app must complete this step on the web app — it cannot be done inside the mobile app.

Once enabled, credits require a prepaid balance. The system includes a $2,000 daily redemption limit and a configurable monthly spending cap; leaving the cap unset means an unconstrained agentic session running at Fable 5 output rates could reach three figures before the session ends. Set a monthly cap before pointing any autonomous agent at the model. An auto-reload option tops up the balance automatically when it drops below a threshold you configure.

Standard Enterprise seats receive no grace period and no included Fable 5 allowance; for those organizations, usage credits must already be enabled or Fable 5 will not function at all. Premium Enterprise seats follow the same July 7 transition as Pro and Max plans.

For teams deciding whether to continue using Fable 5 at all, a few routing principles apply. The model's specific advantages — the 1-million-token context window, the 128,000-token output limit, and the long-horizon agentic coherence that has led enterprises to report compressing months of engineering into days — matter most for large-scale code migrations, multi-day autonomous research sessions, and tasks where reconstructing intermediate state would be more expensive than the Fable 5 price premium. For everything else, Opus 4.8 (inside plan limits at half the rate) and Sonnet 5 (at introductory pricing through August 31) are the routes to cost-controlled agentic work.

The billing event is real. Fable 5's temporary return to subscriptions is not.

Frequently Asked Questions

What happens if I don't enable usage credits before tonight's deadline?

Fable 5 access simply stops. There is no grace period, no automatic fallback to metered billing, and no warning mid-session. If credits are not enabled, any Fable 5 request sent after midnight Pacific Time on July 8 will fail. Opus 4.8, Sonnet, and Haiku remain available inside your normal subscription limits and are unaffected.

Why did the free window end up being only six days instead of the original two weeks?

The original free-inclusion window ran from June 9 through June 22. The US government's export-control directive on June 12 suspended Fable 5 globally before most subscribers had a meaningful opportunity to use it. When export controls were lifted and Fable 5 was restored on July 1, Anthropic provided a compensatory six-day window rather than restoring the full original period, citing capacity constraints. The net result is that the export-control shutdown — triggered by a jailbreak technique that Anthropic's own testing later found could be replicated by every currently available model including Claude Haiku 4.5 — cost subscribers roughly eight days of included frontier-model access.

Does the export-control episode mean Fable 5 could go offline again without warning?

Yes, and that risk is structural rather than temporary. The June 12 shutdown established a legal precedent: the Commerce Department's Bureau of Industry and Security can apply export control authority to a commercially deployed AI model via an Is-Informed Letter, requiring immediate compliance with no advance notice. Anthropic could not comply selectively and had no mechanism to screen users by nationality in real time, so the model was pulled for everyone. That authority still exists. Enterprises building production workflows that depend on Fable 5 should treat administrative revocation — not just outages — as a risk category their business continuity plans now need to address.

When will Fable 5 return to standard subscription plans?

No timeline has been announced. A Claude Code lead engineer said Anthropic aims to restore Fable 5 as a standard part of subscription plans "as soon as capacity allows." Anthropic signed a deal for all of SpaceX's Colossus 1 data center compute — over 220,000 Nvidia GPUs — in May and has additional capacity agreements with Amazon (up to 5 gigawatts), Google and Broadcom (5 gigawatts, coming online from 2027), and Microsoft. Even with that expansion, demand for Fable 5 appears to exceed what Anthropic can serve at flat subscription rates. The return will happen when compute supply catches up; Anthropic has not stated when it expects that to be.