Anthropic’s Fable 5 draws mixed reactions from early users

Anthropic's newly released AI model, Claude Fable 5, is receiving a mixed response from early users, with developers and industry leaders highlighting both major improvements in capability and concerns around cost and access.

Launched alongside Mythos 5, the model is designed to handle complex, long-running tasks such as software development, research and design. According to Anthropic, Fable 5 is a safeguarded version of Mythos, intended for broader public use.

Early reactions on X (formerly Twitter) point to a noticeable jump in performance.


Andrej Karpathy, an AI researcher and former OpenAI and Tesla executive, described the release as "a major-version-bump" and said the model performs strongly on benchmarks and in extended problem-solving sessions.


Similarly, Garry Tan, CEO of Y Combinator, described the model as having "the biggest model energy" he has seen.


Other users highlighted practical applications.

Asaad Mahmood, founder of design agency The Small Square, said he used Fable 5 to design and launch a website, noting that the output showed structured design elements such as hierarchy and spacing typically associated with experienced human designers. He added that the gap between AI-generated and human design work is narrowing.


Marc Andreessen, cofounder of Andreessen Horowitz, pointed to the model's potential in content generation, sharing an example prompt that asks the system to generate large volumes of editorial-style writing with high accuracy.


Use cases highlighted by early testers include real-time software generation during customer calls, automated report and presentation creation, and pixel-accurate design replication.

Deedy Das, an investor and engineer, cited examples ranging from generating consulting-style reports to building interactive 3D environments and UI systems.


Ethan Mollick, a professor at the Wharton School who studies AI adoption, said the model can process large design documents and work for extended periods up to several hours before producing outputs.


Similarly, Dan Shipper, CEO of Every, a media and software company, reported that internal testing showed the model completing large coding tasks autonomously, including clearing production bug backlogs and generating full applications in a single run.

Boris Cherny, a software engineer, said the model demonstrates more structured debugging behaviour, including iterative testing and verification, without explicit prompting.

The limitations

Felix Rieseberg, who leads Claude Code and desktop AI initiatives at Anthropic, said the launch marks a broader transition in how AI systems are used from handling discrete tasks to managing ongoing "responsibilities."


The company has initially rolled out Fable 5 within existing subscription tiers but with usage limits.

Amol Avasare, head of growth at Anthropic, said the model will be included in subscription plans for a limited period approximately two weeks after which access will shift toward usage-based pricing via token consumption.


Some developers note that the model's higher capability comes with increased computational cost.

Shipper said Fable 5 is significantly more expensive than earlier models and consumes substantially more tokens per task, making it better suited for high-complexity workloads rather than routine use.

Bindu Reddy, CEO of Abacus.AI, said internal evaluations suggest performance gains are most evident in a small percentage of highly complex coding tasks.