"White Moonlight" Returns and Instantly Becomes a "Money Pit"? Claude Fable 5 is Subject to Time-Limited Quotas, with Polarized Public Opinions

According to a report by Zhidongxi on July 2nd, early this morning, Anthropic announced that its "mythical" - level Claude Fable 5 was officially unblocked!

Claude announced the return of Fable 5 (Source: X)

It has been 19 days since Fable 5 was taken offline on June 12th due to regulations. The re - launched Fable 5 has no difference in the underlying model capabilities compared to the initial release version. However, for subscribed users, only 50% of the weekly package quota is allocated to Fable 5. Moreover, the monthly subscription method for using Fable 5 will end on July 7th. After these 7 days, users will completely switch to paying by usage credits.

Meanwhile, Claude Fable 5 has upgraded its security risk control to a more strict level. It is prone to mis - intercept normal development requirements. Once triggered, it will refuse to answer and automatically degrade to Opus 4.8.

After the return of Fable 5, the word - of - mouth is polarized. Many users' actual tests show that the model's performance has not declined and it can efficiently complete complex tasks such as design and 3D game development. However, some users complain that it consumes a large number of tokens. Even a simple question can quickly deplete the package quota. At the same time, the risk control judgment is too broad. Normal requirements such as code and medical analysis are often mis - intercepted, and the model will be gradually degraded to a lower - level model.

Claude Fable 5 is a Mythos - level flagship model launched by AI unicorn company Anthropic on June 9th. Its code processing, complex task, and visual reasoning abilities are superior to those of the Opus series. When the model was first launched, subscribed users could use it for free in full. It was taken offline globally due to U.S. export regulations only three days after its launch and was blocked for nearly 19 days. After the regulations were lifted on June 30th, Fable 5 was re - launched yesterday.

01. 50% Quota for 7 - day Limited Time, Purchase Usage Credits to Continue Using

Along with the return of Fable 5 comes a promotional usage instruction. In this instruction, Claude clearly states that during the event period from July 1st to July 7th, subscribed users can use up to 50% of the weekly package quota to call Claude Fable 5 without additional payment.

After exceeding this quota, users must top up the pay - as - you - go balance to continue using Fable 5. They can also switch to other models and use the remaining quota in the package to continue operating.

7 - day promotion instructions for Fable 5 (Source: Anthropic)

It should be noted that Fable 5 will occupy the regular weekly quota of the package. Compared with other Claude models, it consumes the quota faster. Under the same task, the consumption is about twice that of Opus 4.8, and it is very easy to quickly exhaust the weekly quota.

If users use up the 50% free share in advance, they have two options: one is to activate the usage credits balance and pay separately to continue using Fable 5; the other is to switch to other models and consume the remaining subscription quota to complete the work.

That is to say, within these 7 days, subscribed users have 50% of the quota to use Claude Fable 5. After 7 days, users who want to use Fable 5 can only do so by activating usage credits and paying extra.

02. Risk Control Upgraded, Regular Questions May Also Be Intercepted

In terms of security design, Claude Fable 5 adopts a more strict "defense in depth" mechanism compared to before. It reduces the risk of abuse by training the model to reject high - risk requests, introducing an automated classifier to identify potential dangerous operations, and blocking the model output when necessary.

Among them, the classifier system will judge in real - time whether the user's request involves high - risk behavior during the interaction process. Once the risk control rules are triggered, it will directly intercept the answer or automatically switch to a lower - level model such as Opus 4.8 to reply.

Security boundary of Fable 5 (Source: Anthropic)

To further reduce the risk, the security system of Fable 5 sets a wider "security boundary". This means that some originally normal requests may be misjudged as high - risk, resulting in a refusal to answer. Claude said it will optimize this within a few weeks.

Claude's explanation for the increased risk control of Fable 5 (Source: X)

03. Divergent Return Experience: High Token Consumption and Strict Risk Control Are Two Major Complaints from Users

The unblocking of Claude Fable 5 is like the return of a "white moonlight". Many users have high expectations for its re - launch. One user asked in the Claude comment section, "How many people are as excited as the person in the picture?" Another user said that he was the happiest person in the world after Fable 5 returned.

Users' happiness about the return of Fable 5 (Source: X)

Many users have tested the returned Fable 5. They believe that the capabilities of Fable 5 are still the same as the initial version and have not been weakened. One user said that after the unblocking of Fable 5, its reasoning and design abilities are exactly the same as before it was taken offline. It can efficiently complete a large amount of work within half a day, and the output quality is irreplaceable. At the same time, the user suggested seizing the 7 - day limited - time subscription period to experience it.

Users' experience evaluation after the return of Fable 5 (Source: X)

Many users have shared cases to prove the strong strength of Fable 5. A developer used only a short prompt to let the model independently produce a complete and runnable 3D strategy game. This creation consumed $173 (about 1,174 RMB) worth of tokens.

A game created by Fable 5 shared by a user (Source: X)

However, some users are surprised by the speed at which Fable 5 consumes tokens. They think it is too expensive and have questions about what to do after the 7 - day promotion ends. One user shared that in the Max package, each time Fable 5 answers a simple question, it will consume 2% of the 5 - hour quota.

Users' opinions on the high consumption of Fable 5 (Source: X)

A user who subscribed to a $200 (about 1,357 RMB) package only asked one question. Fable 5 didn't give an answer but consumed 25% of the quota.

Users' opinions on the high consumption of Fable 5 (Source: X)

In addition to the high cost of token consumption, some users complain about the one - size - fits - all risk control restrictions of the new version. One user only inquired about questions related to code synchronization tools and triggered the security interception prompt of Fable 5.

Users' views on the increased risk control of Fable 5 (Source: X)

Another user just asked Fable 5 to interpret a blood report and triggered a series of model downgrades. Both Fable 5 and Opus 4.8 judged that the content had a security risk, and finally the user could only use Sonnet 4.6.

Users' views on the increased risk control of Fable 5 (Source: X)

04. Conclusion: Tightened Risk Control, Exit of Subscription, Compromised Survival of Fable 5

The tortuous launch, blocking, and return of Fable 5 are essentially a compromise and balance made by top - level large models with strong capabilities between regulatory constraints and commercial benefits. To meet the requirements of export regulations, Anthropic can only first limit the model's capabilities through broad - based security risk control. At the same time, it uses the 7 - day 50% quota welfare to buffer the gap in users' experience.

After the 7 - day discount period ends, it will completely break away from the subscription system and switch to pure pay - as - you - go. For developers and enterprise users, they can experience the code and complex task processing capabilities of Fable 5 through the limited - time window in the short term. In the long term, they need to pay separately for high - order AI capabilities. This will also become a common trend in the commercialization of the next - generation flagship large models.

This article is from the WeChat official account "Zhidongxi" (ID: zhidxcom). The author is Qiezi, and the editor is Cheng Qian. It is published by 36Kr with authorization.