Anthropic's latest AI model, Claude Fable, described as the company's most powerful model yet, has now been made available to a wider group of users. But initial reactions from developers and AI users suggest that the launch has not lived up to expectations.
When the news came that the restrictions around Claude Fable had been lifted, many were eagerly waiting for the model to become available in Claude Code. Several users had also topped up their usage credits in anticipation of the model being launched as a purely usage-based service.
To the surprise of many, it was Claude Fable also available to users with the $100 per month Max subscription. At the same time, several restrictions were introduced that affect how the model can be used.
According to Anthropic Fable is included in the Max, Pro and Team subscriptions, but usage is limited. For example, the model can only be used for up to 50 percent of the weekly usage quota. After July 7, Fable is also expected to transition completely to a credit-based system where users pay based on actual usage.
Users report degraded performance
The biggest criticism, however, is not about the limitations of the subscriptions but about the model's perceived performance. Within the AI community, several users describe Claude Fable as feeling "nerfed," meaning that the model delivers worse results than previous versions.
On Reddit reports Several users report that the relaunched model is significantly more often steered over to older models due to the new security systems.
A user writes:
“The new guardrails stop far too many requests and fall back to Opus 4.8. This is not the model that was previously blocked.”
Similar reports also come from Claude Code, where several developers experience that the model is shifting to Opus 4.8 significantly more often than expected.
A user states that Claude Fable did not even allow a search for unused code before the model automatically switched to Opus. Another describes the transition as very clear because Claude informs the user of the model change when it occurs.
Coding tasks seem to trigger the security features
Several developers also describe problems with low-level programming work. Tasks involving C, C++, Rust, Win32 API, memory management or files with words like “security”, “vulnerability”, “insecure” or “hook” are reported to trigger security features or automatic model switches more often.
This means that Claude Fable can still deliver very high performance when actually used, but that the new version seems significantly more sensitive to certain types of instructions, project files, and security-related tasks.
It's probably about stricter security measures
At the same time, there is no indication that the AI model itself has been technically degraded. A more likely explanation is that Anthropic has implemented significantly stricter security mechanisms that in some cases are activated even for relatively harmless tasks.
Observations show that Claude Fable sometimes switching to Opus 4.8 even when the task does not seem to pose any obvious security risk.
Anthropic has previously stated that the company's updated protection system is based on a large safety margin to minimize the risk of abuse. That strategy can also lead to more so-called false positives, where legitimate data is unnecessarily restricted.
The company has yet to comment on the reports circulating about false positives or the perceived performance degradation. If the issues prove widespread, it is likely that Anthropic will adjust the security mechanisms in an upcoming update.
Swedish
Danish
Norwegian
Finnish
English
German
Dutch