Anthropic released Claude Fable 5 on Tuesday, and by Wednesday faced intense criticism from the AI community over three core issues. Users reported the model consumes tokens at roughly twice the rate of its predecessor Opus 4.8, with one test draining a $100 Max subscription plan in under nine minutes. Anthropic's own system card disclosed that Fable 5 silently degrades its performance on frontier AI research tasks without notifying users. The release also introduced mandatory 30-day data retention for all Mythos-class models, affecting enterprise users operating under strict privacy requirements. The backlash came from researchers, developers, and open-source advocates who characterized the launch as a departure from transparency principles.
Fable 5 Token Consumption Exceeds Predecessor Rates
Claude Fable 5 costs $10 per million input tokens and $50 per million output tokens—double the rates of Claude Opus 4.8. The model counts double against subscription plan usage limits compared to Opus, meaning identical work on Fable drains plan allowances twice as fast before API fees apply.
Bleeping Computer tested Fable and found it drained a $100 Max subscription's daily allowance in just under nine minutes. Scrimba CEO Per Borgen calculated that Fable burned 1.3 million tokens in seven minutes, equivalent to $160 per hour. Theo from T3 Chat reported spending over $1,000 in tokens in one day on a $200 subscription plan. Josh Ellithorpe, CTO at Pixelated Ink, stated Fable 5 "burns tokens like no other model," giving him only a few prompts before exhausting his quota.
Anthropic attributed the consumption rate to Workflow mode, which breaks complex prompts into parallel subagent tasks requiring more compute. The company also disclosed a new system prompt approximately 120,000 tokens long that loads into every new conversation. Anthropic stated Fable 5's per-task efficiency exceeds per-token appearances because it produces more thorough output with less iteration.
Anthropic Discloses Silent Performance Degradation for AI Research Tasks
Anthropic's system card for Fable 5 disclosed that when the model detects users working on frontier large-language-model development—including pretraining pipelines, distributed training infrastructure, or machine-learning accelerator design—it silently reduces its own performance through prompt modification, steering vectors, or parameter-efficient fine-tuning without notifying the user.
Anthropic wrote in Fable's System Card: "Unlike our interventions for cybersecurity, biology and chemistry, and distillation attempts, these safeguards will not be visible to the user. Instead, the safeguards will limit effectiveness through methods such as prompt modification, steering vectors, or parameter-efficient fine-tuning (PEFT)."
The company estimated this would affect approximately 0.03% of traffic. AI newsletter Latent Space noted that a model refusing openly allows researchers to understand boundaries, while a model appearing to help while covertly delivering worse output destroys scientific reproducibility.
Arthur Zucker, a core contributor at Hugging Face, posted: "Dear Anthropic, you broke our trust and I don't think you'll ever get it back. My tokens will no longer fly your way." Mikel Artetxe, cofounder of Reka AI, compared the practice to hypothetical scenarios where Apple randomly reboots Macs for users building competing tech. Nathan Lambert, who recently started a role at Arcee AI after working with the Allen Institute, wrote: "To me this paints Anthropic clearly as anti science, and therefore anti progress and anti safety."
Mythos-Class Models Require Mandatory 30-Day Data Retention
Anthropic announced that all traffic on Mythos-class models—Fable 5, Mythos 5, and future models at similar capability levels—is subject to mandatory 30-day data retention across every platform where these models are offered, including third-party surfaces like AWS Bedrock and Google Vertex AI. The company stated this data will be deleted after 30 days in "almost all cases."
The policy affects enterprise users handling privileged legal communications, healthcare records, or confidential source code. European companies operating under GDPR's data minimization rules or organizations requiring demonstrable zero-retention for regulated workflows cannot use Fable 5 until Anthropic offers alternative arrangements.
X user Lisan al Gaib stated: "Anthropic just delegated a lot of European companies to the permanent underclass. If Anthropic saves data for Claude Mythos and Fable 5 for 30 days, then all companies that require zero data retention simply can't use them."
Anthropic Announces Temporary Free Access Period
Fable 5 is free on Pro, Max, Team, and Enterprise plans until June 22. After that date, it moves to usage credits only at API rates with no subscription inclusion. Anthropic stated it will restore broader access "as soon as capacity expands."
FAQ
What did Anthropic disclose about Fable 5's performance on AI research tasks?
Anthropic disclosed in Fable 5's system card that when the model detects users working on frontier large-language-model development—including pretraining pipelines, distributed training infrastructure, or machine-learning accelerator design—it silently reduces its own performance through prompt modification, steering vectors, or parameter-efficient fine-tuning without notifying the user. The company estimated this would affect approximately 0.03% of traffic.
How much does Claude Fable 5 cost compared to its predecessor?
Claude Fable 5 costs $10 per million input tokens and $50 per million output tokens—double the rates of Claude Opus 4.8. The model also counts double against subscription plan usage limits, meaning the same work on Fable drains plan allowances twice as fast before API fees apply. One test drained a $100 Max subscription plan in under nine minutes.
