Skip to content
NewsProduct launch

Fable 5 leaves Anthropic subscriptions today as per-token billing begins at $10/$50 per million tokens

· by Pondero Newsdesk

The short version

Starting July 8, 2026, Claude Fable 5 exits Pro, Max, Team, and select Enterprise subscription plans. Usage credits at $10 input / $50 output per million tokens become the only access path, with no timeline given for restoration to standard tiers.

Fable 5 leaves Anthropic subscriptions today as per-token billing begins at $10/$50 per million tokens

The free-inclusion window for Claude Fable 5 expired at the end of July 7. Every Pro, Max, Team, and premium Enterprise subscriber who wants to keep using the model must now enable usage credits in the Claude Console. The rate: $10 per million input tokens and $50 per million output tokens, per Anthropic's redeployment announcement. That is exactly double Opus 4.8, and it is the highest per-token rate Anthropic has ever published for a generally available model.

What happened and when

Fable 5 launched on June 9 as Anthropic's most capable publicly accessible model, initially bundled into paid subscription tiers through June 22. That first free window never delivered its full run. Three days after launch, on June 12, the US Commerce Department imposed export controls on Fable 5 and Mythos 5, requiring Anthropic to suspend both models globally within roughly 90 minutes because the company had no mechanism to verify nationality at consumer scale in real time. The 19-day blackout ran through June 30.

Commerce Secretary Howard Lutnick withdrew the export-control requirement on June 30. Anthropic restored Fable 5 access on July 1 for users on the Claude Platform, Claude.ai, Claude Code, and Claude Cowork, with a compressed six-day grace period running through July 7. Per Anthropic's redeployment post, Pro, Max, Team, and select Enterprise plans received Fable 5 for up to 50% of weekly usage limits during that window at no additional cost.

That grace period ended at midnight. Standard Enterprise seats received no inclusion window at all: Fable 5 has required enabled usage credits for those users from day one. Third-party cloud providers including AWS, Google Cloud, and Microsoft Foundry remain in a rolling restoration process, with no confirmed completion date announced by Anthropic.

The billing math for agentic workloads

The headline $50-per-million-output rate becomes significant fast in real-world agent loops. Transformer models like Fable 5 are stateless across API calls: every piece of accumulated context, the project plan, tool definitions, and prior turn summaries, must be passed in explicitly with each new request. Per TechTimes, a $100 Max subscription was drained in under nine minutes during heavy Fable 5 testing. Scrimba's CEO logged 1.3 million tokens in seven minutes, a pace equivalent to roughly $160 per hour at the published rate. Independent tracking of production deployments found some agent loops running $200 to $400 per hour during intensive sessions.

There are three cost levers teams can apply immediately. Prompt caching cuts input costs by 90%, from $10 to $1 per million tokens, on repeated content such as system prompts and stable file context. The Batch API halves both rates ($5 input, $25 output per million tokens) for any workload that does not require real-time responses. Task routing, reserving Fable 5 for genuinely complex multi-step planning and running Sonnet 5 or Opus 4.8 on everything else, is the operational control that matters most at scale.

One additional complexity: Fable 5's new safety classifier, added alongside the July 1 relaunch to address a documented Amazon-researched jailbreak technique, carries a known false-positive rate on routine coding and debugging tasks. Per TechTimes, citing BridgeMind testing of a TypeScript debugging benchmark, only 3 of 12 tasks ran to completion on Fable 5 after the relaunch. The other 9 were rerouted to Opus 4.8, billed at Opus rates. Teams that built pipelines expecting Fable 5 outputs may receive Opus 4.8 responses on a meaningful share of coding requests.

What changes today for each plan type

The access picture differs by plan tier. Pro, Max, and Team subscribers who do not enable usage credits lose Fable 5 access entirely today. No fallback to Fable 5 exists within the flat monthly fee; the model simply becomes unavailable when credits run out or are not activated.

Premium Enterprise seats face the same cutoff. Standard Enterprise seats never had an included Fable 5 window: for those organizations, Fable 5 has never been accessible unless an administrator explicitly enabled usage credits from July 1 onward. Per the DigitalApplied pricing breakdown, standard Enterprise seats also carry no Zero Data Retention option for Fable 5 or Mythos 5, both of which are designated Covered Models with a 30-day data retention period. Procurement teams building on Fable 5 for sensitive workloads need to account for this before deployment.

Anthropic said in the June 30 announcement that Fable 5 will return to standard subscription tiers "as soon as capacity allows," attributing the exclusion to unexpectedly high demand following the model's return. No timeline was published.

Why it matters for teams building on Fable 5 today

The practical action is narrow and time-sensitive. Any team that did not enable usage credits before today should log into the Claude Console, enable credits, and set a spending cap to avoid uncapped exposure. Without active credits, Fable 5 requests return errors rather than falling back to another model silently.

For budget planning, the $10/$50 rate makes Fable 5 roughly five times more expensive per token than Sonnet 5 at its current introductory rate of $2/$10 per million tokens. For tasks where Fable 5's long-horizon coherence in multi-step agentic work genuinely matters, the premium may justify itself. For bulk processing, code generation, or document summarization, Sonnet 5 or Opus 4.8 produce comparable results at a fraction of the cost.

The data retention constraint is a harder blocker for some organizations than the price. Teams running compliance-sensitive workflows, where data retention periods are contractually limited, cannot currently opt out of the 30-day Covered Model policy for Fable 5. That constraint does not apply to earlier Claude models, which remain available under standard API terms.

Anthropic's commitment to restore Fable 5 to subscription tiers has no attached milestone. Capacity demand and the government's ongoing review of the jailbreak framework Anthropic proposed in its June 30 post are the two variables most likely to influence timing. Both remain open as of today.

What to watch next

Three signals are worth tracking. First, whether Anthropic announces a capacity timeline for returning Fable 5 to standard subscription tiers, and under what conditions. Second, whether the false-positive rate on the new safety classifier improves in a near-term patch, which would make Fable 5's $50 output rate more reliably purchased rather than silently rerouted. Third, whether the Glasswing-partner jailbreak severity framework that Anthropic, Amazon, Microsoft, and Google began drafting in the June 30 announcement advances to a published standard, which would shape how future frontier model releases are handled at the subscription layer.

Sources