On June 9, 2026, Anthropic launched Claude Fable 5 — the public version of its Mythos-class capability — and included access on Pro, Max, Team, and seat-based Enterprise plans at no extra cost through June 22. As of today, June 23, that inclusion has ended. Continued Fable 5 use on those plans now draws from prepaid usage credits billed at API rates: $10 per million input tokens, $50 per million output tokens. That is double the price of Claude Opus 4.8.
The billing change is immediate and specific. What it signals about how frontier AI model access will be priced as models advance is worth more attention than the line item itself.
stateDiagram-v2 direction TB state "Jun 9-22: Fable 5 on plan (free window)" as Free state "Jun 23+: Usage credits required at API rates" as Credits state "Evaluate: does this task need Fable 5?" as Eval state "Opus 4.8 — standard plan rates" as Opus state "Fable 5 via credits — approved spend" as Fable [*] --> Free Free --> Credits: June 22 cutover Credits --> Eval Eval --> Opus: Routine or single-turn work Eval --> Fable: Long-horizon agentic task Opus --> [*] Fable --> [*]
The Rundown: Two Weeks Free, Then Credits
Anthropic released Fable 5 on June 9, 2026, as the publicly accessible version of its Mythos architecture. Claude Mythos 5, the less restricted version, is being deployed separately through Project Glasswing with US government partners for cybersecurity and critical infrastructure work. Fable 5 is the version available to enterprise and API customers.
During the June 9–22 free window, Fable 5 access on seat-based plans was not entirely without cost. Per Anthropic’s documentation, Fable 5 sessions counted at roughly twice the usage rate of Opus 4.8 against plan limits — meaning teams running heavy Fable 5 workloads during the free window were burning through plan allowances approximately twice as fast as they may have realized.
As of June 23, usage credits billed at $10 per million input tokens and $50 per million output tokens apply. That is double the Opus 4.8 API rate on both input and output. Fable 5 and Mythos 5 launched at the same price point. Anthropic has not announced when or whether plan inclusion will be restored; capacity constraints on frontier-class inference are the stated reason for the June 22 cutoff.
For Engineers: Where the Capability Premium Is Actually Justified
Fable 5’s meaningful capability advantages over Opus 4.8 are concentrated in a specific category of work. Extended agentic tasks — multi-hour sessions, large codebase migrations, deep research tasks where the model must maintain coherence over very long contexts — are where Fable 5’s architecture earns the premium. For those workloads, the quality difference is measurable.
For standard work — document processing, single-turn question answering, code review of individual functions, structured output generation, routine classification — the performance gap between Fable 5 and Opus 4.8 is narrow enough that Opus 4.8 is the right default at half the price. Most teams evaluating Fable 5 during the free window discovered this: the model is impressively capable across a wide range of tasks, but the tasks that genuinely require it are a subset of total usage.
The engineering implication of the pricing change is direct. If your team’s Fable 5 usage during the free window was not dominated by long-horizon agentic work, you are now paying a 2x premium for a marginal quality improvement on tasks Opus 4.8 handles well. Auditing which task types your team directed at Fable 5 and setting routing logic accordingly is the right near-term action.
For Business Owners: What This Billing Change Reveals About AI Cost Planning
The transition from flat subscription access to usage-credit billing is predictable from a provider perspective and frequently surprising from a customer perspective. Anthropic is capacity-constrained on Fable 5. Frontier-class inference is expensive to run. A flat subscription model means Anthropic absorbs cost overruns from heavy users. Usage credits shift that dynamic and create a direct cost signal where none previously existed.
The immediate budget question for enterprise teams: how much Fable 5 usage did your engineers run during the free window, and what would that have cost at API rates? If the answer is not readily available, you do not have the usage visibility needed to manage AI costs as model capabilities advance.
A single engineer running intensive Fable 5 sessions — say, a multi-hour code migration at 500,000 tokens — would incur roughly $30 per session at the new rates, compared to $15 on Opus 4.8. Across an engineering team running several such sessions per week, the delta is a real budget line. Across an organization that defaulted to Fable 5 out of habit rather than need, it can be a significant and unplanned spend.
The pattern this establishes is the more important business lesson. As Anthropic continues to push frontier capability, the gap between what is included in a subscription and what requires API-rate credits is likely to expand over time. Organizations with usage governance — defined routing policies, cost attribution by team, visibility into model-tier usage — will absorb each transition without friction. Organizations without that infrastructure will cycle through billing surprises as the tier structure evolves.
My Take: The Free Window Was a Structured Evaluation, and Most Teams Did Not Treat It That Way
Two weeks of free Fable 5 access is long enough to assess whether the capability justifies the cost at your specific workloads. Most organizations used it as a benefit rather than as an evaluation. That is understandable — running a structured model evaluation while also doing actual work is overhead that rarely gets prioritized when the model is simply available at no incremental cost.
The result: teams now face usage-credit billing for a model tier they did not formally evaluate, on workloads they did not formally assess. Some will default to Fable 5 because that is what they got used to. Those teams will find elevated costs in their July billing that were not in any plan.
The practical correction is clear. Identify which task types in your current AI workload genuinely require long-horizon autonomous operation — the things Opus 4.8 fails on in ways that cost you real time or rework. Those are the Fable 5 cases. Route everything else to Opus 4.8 or Sonnet by default. Document that routing decision as a policy so it is not an individual engineer choice made on instinct.
Anthropic has indicated the plan inclusion may be restored when capacity allows. Treat that as uncertain and plan the credit model as the steady state. If free access returns, treat it as a windfall rather than an expectation. The pricing signal is real regardless of when or whether that changes.
