Claude Fable 5's complimentary access window ended last night. Starting June 23, every request draws from usage credits at $10 input and $50 output per million tokens, double the cost of Opus 4.8. Here is what that means for teams that built on it.
For the agency team that replaced two researchers with a Fable 5 workflow that processes 400 pages of competitive material and returns a structured briefing in 90 minutes, today is the day that workflow stops being "included." Anthropic launched Claude Fable 5 on June 9 with 13 days of complimentary access for Pro, Max, Team, and Enterprise subscribers. That window closed last night. Starting June 23, every Fable 5 request draws from purchased usage credits at $10 per million input tokens and $50 per million output tokens. That is exactly double the cost of Opus 4.8. If your team built on Fable 5 during the free period and has not audited its integrations, the interesting conversation today is with your credit balance, not your model.
The math is not complicated but it is easy to underestimate. Fable 5 uses a newer tokenizer that produces roughly 30 percent more tokens from the same text than older Claude models. A workflow that outputs 100,000 tokens per day costs about $5 in output credits at Fable 5 rates, or roughly $150 a month on top of whatever subscription you already pay. For a team doing that volume daily, the annual cost is $1,800. That number survives most finance conversations. What does not always survive is the 10x version of it, which is what you get from a longer agentic chain you did not cost out before it ran.
The timeline adds a wrinkle. Anthropic designed the 13-day window to give teams time to evaluate Fable 5, benchmark it against Opus 4.8, and make a routing decision before billing began. Then on June 12, three days after launch, both Fable 5 and Mythos 5 went offline globally after a US Commerce Department export control directive. Access came back around June 18. Teams that built on Fable 5 during the first three days and then had access cut got roughly four to five days of actual evaluation time, not 13. Today is the first real day of production billing on workflows some teams barely had a chance to stress-test.
That compressed window matters operationally. If your integration uses model aliases or "latest" style routing rather than an explicit model ID, and your alias resolved to Fable 5 during the complimentary window, it is still resolving to Fable 5 now and drawing credits at $10/$50 per million. The Fable 5 model ID is claude-fable-5. Any call that does not specify it explicitly should be checked before the end of business today.
The fallback question is cleaner than it might seem. Opus 4.8 and Fable 5 share the same 1 million token context window and the same 128,000 token maximum output per request. The API surface is compatible. Switching is a one-line change to the model identifier. Fable 5 leads on very long agentic workflows, complex reasoning chains, and tasks that require its specific depth of judgment. For campaign writing, research synthesis, content generation, and most operations workflows, the benchmark gap between the two models is real but not always worth the 2x price difference. That is not a general rule. It is a question you need to answer for your specific use case with your specific token volumes.
There is also an honest caveat about what "double the cost" means in practice for agentic work. A workflow that runs 10 million output tokens in a month costs $500 in Fable 5 credits. The same workflow through Opus 4.8 costs $250. That delta is predictable and modelable, but only if you know your token volume. Many teams do not. The Claude billing page shows monthly token consumption by model. This week is the week to look at it before the credits move.
One more thing worth noting: Anthropic has committed to restoring Fable 5 as a standard flat-rate feature on subscription plans once compute capacity allows, but has not announced a date. The implicit premise of the launch, that subscribers would eventually get the best model included in what they already pay, is still the stated intent. It is just deferred. Whether that deferred promise factors into your team's near-term spend decisions is a judgment call, but it should be an explicit one.
The broader shift underneath all of this is structural. For the last two years the operating assumption across most AI-forward teams was: try everything at low cost, optimize later. That assumption was always provisional. Flat-rate access to frontier models was a customer acquisition posture, not a sustainable pricing model. The Fable 5 credit cliff is the industry acknowledging that out loud. The teams that built workflows during the free window without benchmarking the cost did not get tricked. They got a trial. Today the trial ends and the bill arrives, which is exactly what trials are designed to do.