Claude Fable 5 and Mythos 5 Release

Anthropic has released Claude Fable 5 and Claude Mythos 5, marking a significant leap in AI capability, particularly in software engineering, scientific research, and complex analytical tasks. While Fable 5 is generally available, Mythos 5 is restricted to a trusted access program for cyberdefenders and life sciences researchers.

High-Level Capabilities and Performance

Claude Fable 5 is designed for long-horizon, autonomous tasks that previously required extensive human steering. It demonstrates state-of-the-art performance across nearly all tested benchmarks, with its lead over previous models increasing as task complexity grows.

Software Engineering and Coding

Fable 5 shows a marked improvement in agentic coding and autonomy.

Codebase Migrations: In early testing, Stripe reported that Fable 5 completed a codebase-wide migration in a 50-million-line Ruby codebase in a single day—a task that would have taken a human team over two months.
Production Standards: On Cognition's FrontierCode evaluation, Fable 5 scored the highest among frontier models, demonstrating a superior ability to meet high-quality production code standards.
Efficiency: Early user feedback indicates the model is more token-efficient, often delivering results in fewer turns and producing more surgical, maintainable diffs.

Scientific Research and Drug Design

Through Claude Mythos 5, Anthropic has achieved breakthroughs in autonomous scientific discovery:

Protein Design: Internal experts reported a 10x acceleration in the drug design process. Mythos 5 matched or beat skilled human operators in selecting binding sites and running protein design tools without human assistance.
Molecular Biology: The model consistently produces novel scientific hypotheses, with scientists preferring Mythos's hypotheses over Opus-class models ~80% of the time.
Genomics: Mythos 5 conducted a week of autonomous genomics research, training a custom machine learning model that outperformed a recent study published in Science despite being 100 times smaller.

Vision and Analytical Reasoning

Fable 5 introduces new state-of-the-art capabilities in vision and complex reasoning:

Vision Tasks: The model can rebuild a web app's source code from screenshots alone and extract precise numbers from scientific figures.
Knowledge Work: On Hebbia's Finance Benchmark, Fable 5 achieved the highest score of any model in document-based reasoning and chart interpretation.
Memory: In tests playing Slay the Spire, persistent file-based memory improved Fable 5's performance three times more than it did for Opus 4.8.

Safety Framework and Safeguards

Because Mythos-class models pose significant risks in cybersecurity and biology, Anthropic has implemented a tiered safety system.

The Fallback Mechanism

Fable 5 uses separate AI classifiers to detect potential misuse. When a request is flagged as relating to cybersecurity, biology, chemistry, or distillation, the system automatically routes the query to Claude Opus 4.8 instead of Fable 5. This ensures the user receives a response from a highly capable model rather than an outright refusal, though Anthropic notes these safeguards are tuned conservatively and may trigger false positives in less than 5% of sessions.

Specific Risk Areas

Cybersecurity: Classifiers prevent Fable 5 from performing agentic hacking tasks, such as reconnaissance and lateral movement.
Biology and Chemistry: Due to the dual-use nature of biological research, most requests in these fields fall back to Opus 4.8 to prevent the creation of dangerous pathogens.
Distillation: To prevent the extraction of capabilities to train competing models, distillation attempts are also routed to Opus 4.8.
LLM Development: The model is intentionally limited in its effectiveness for requests targeting frontier LLM development (e.g., pretraining pipelines) to prevent the acceleration of competing models.

Pricing and Availability

Claude Fable 5 is available via the Claude API and subscription plans.

Pricing: $10 per million input tokens and $50 per million output tokens.
Subscription Access: Fable 5 is included in Pro, Max, and Team plans at no extra cost until June 22. After this date, usage will require usage credits, with a plan to restore it as a standard part of subscriptions once capacity allows.
Data Retention: A new policy requires 30-day retention for all traffic on Mythos-class models for safety and attack defense purposes, though this data is not used for training.

Community Insights and Critical Feedback

User discussions on Hacker News highlight a divide between those impressed by the model's reasoning and those frustrated by its constraints.

"I've spent some time with this now... it's a beast. I'm throwing some VERY difficult problems at it—things I've been dragging my heels on for months—and it's crunching through them very happily."

However, critics have raised several concerns:

Over-aggressive Filtering: Many users reported that benign tasks—such as writing GPU drivers or analyzing DNA results—triggered the cybersecurity or biology filters, forcing a fallback to Opus 4.8.
Anti-competitive Behavior: Some users suggest that by restricting the model's ability to help with LLM development and biology, Anthropic is using "safety" as a cover to protect its own commercial interests in those fields.
Cost Concerns: The high token cost of Fable 5 compared to previous models has led some developers to question the economic viability of using it for daily production workflows.