Claude Fable 5: Four Reasons Anthropic Keeps Giving Away Its Most Expensive Model
Anthropic has extended free Claude Fable 5 access to 19 July, the second extension in under a week. Here is what the delay tells buyers about its position.
The Bright Recap
Free access to Claude Fable 5 on paid Anthropic plans now runs to 19 July 2026. The window was set to close on 7 July, was moved to 12 July, and was extended again on 12 July. Once it ends, the model costs $10 per million input tokens and $50 per million output tokens in prepaid credits.
To know more about this topic, read our related articles:
- The conditions for the model's return
- How OpenAI priced its answer
- AI's hidden cost problem
- Financial technology explained
Bright Answers
How long is Claude Fable 5 free?
Until 19 July 2026, on Pro, Max, Team and premium Enterprise seats, for up to half of a plan's weekly usage. Free-tier accounts, standard enterprise seats and direct API customers were never included.
Why does Anthropic keep extending it?
Officially, compute capacity. The extensions also coincide with two rival flagship launches, with Fable 5 carrying the highest cost per finished task of any leading model, and with a safety filter that hands some requests to a cheaper Claude model.
Every free trial has an end date, and when the seller keeps moving it, the reason belongs to the seller. Anthropic has now pushed back the day it starts charging for Claude Fable 5 twice in under a week. Free use of its most capable and most expensive model was due to stop on 7 July 2026, then on 12 July, and it now runs to 19 July. The company has given no reason for either move.
What arrives on 19 July is a price. Fable 5 becomes payable in prepaid credits at $10 per million tokens of input and $50 per million tokens of output, tokens being the word fragments these systems bill by, and that is double the rate of the Claude model sitting directly beneath it. Four things happened around the decision, and each is a reason a company might prefer to keep the meter switched off a little longer.
What ends on 19 July
Subscribers on Pro, Max, Team and premium Enterprise seats can currently spend up to half their weekly allowance on Fable 5 at no extra cost. Free accounts, standard enterprise seats and direct API customers were never part of it. The model is only available to any of them because it came back under conditions, having been switched off by a United States export-control order on 12 June, three days after it launched.
Reason one: two rivals shipped in the same week
SpaceXAI released Grok 4.5 on 8 July, and OpenAI made GPT-5.6 generally available on 9 July, holding its price flat while claiming the new flagship finishes work using far less of what it bills for. Anthropic extended its free window three days after that. A developer holding a free week of the best Claude model is a developer who does not spend that week setting up an account somewhere else.
Reason two: the price stops looking reasonable the moment anyone compares
The independent evaluator Artificial Analysis measured what a single finished coding task costs rather than what each system charges per token. Grok 4.5 came out at $2.49 per task, GPT-5.5 at $5.07, and Fable 5 at $11.80. Appetite explains the gap, with Fable 5 getting through an average of 7.2 million tokens to complete a task against 1.9 million for Grok. Switching on a $50 meter in the middle of that comparison invites every customer to run the arithmetic themselves.
Reason three: the model does not always answer
Fable 5 returned on 1 July with a new safety filter attached as a condition of its reinstatement. Anthropic's own redeployment notice states that the filter flags harmless requests more often during routine coding and debugging, and that those requests are handled by the cheaper Claude Opus 4.8 instead, with the customer told once it has happened. A company that cannot promise which model will respond is in a poor position to start charging $50 per million tokens for one of them.
Reason four: the one Anthropic actually gives
On the underlying arrangement, the company has been consistent. It says the credit system is temporary and that Fable 5 returns to standard subscriptions once there is enough computing capacity to serve it. That explanation is plausible and costly at the same time, because every free week is computing power Anthropic pays for and bills to nobody, at a moment when the cost of running these models is the question hanging over the entire sector.
What a buyer should take from this
The cheapest option is not automatically the right one. Artificial Analysis still ranks Fable 5 above Grok 4.5 on its coding agent index, and it found Grok's rate of confident false answers climbing from 25% to 54% against the previous Grok release, even as the model's accuracy improved. Anyone buying software in financial technology already asks the only question that settles this, which is whether the invoice tracks what the tool does or what the seller says it does.
A free trial gets extended when the seller is not yet confident the customer will pay at the end of it. Anthropic has extended this one twice in under a week.
Editor's note
Every piece published on The Bright Minded goes through careful verification, but mistakes can happen. If you spot an error, have additional information, or want to flag anything, write to rosalia@thebrightminded.com.