Anthropic Launches Claude Fable 5 with Enhanced Safety Controls

March 25, 2025 — Anthropic has launched Claude Fable 5, a generally available Mythos-class model featuring new safety controls that handle longer, more complex tasks than prior Claude models. The release comes with safeguards that route certain cybersecurity, biology, chemistry, and distillation requests to Claude Opus 4.8 instead.

Immediate Details & Direct Quotes

Claude Fable 5 is now available to users through Claude products and the Claude API, with developers accessing it via the claude-fable-5 API identifier. According to Anthropic, the model performs strongly in software engineering, knowledge work, vision, and scientific research, with its largest lead on longer and more complex tasks.

The company said Fable 5 can work autonomously for longer periods than earlier Claude models and stay focused across millions of tokens in long-running tasks. In software testing, Stripe reported that Fable 5 completed a large Ruby migration in one day—a task the company said would have taken a team more than two months by hand.

“The safeguards may catch harmless requests because we tuned them conservatively,” Anthropic stated. The company wants to reduce false positives after launch and added new classifiers to detect potential misuse and jailbreak attempts. These systems prevent Fable 5 from responding directly to flagged requests.

Anthropic also cited strong results on finance, vision, memory, and scientific research tasks, though noted the release required extra controls due to the model’s capabilities.

Market Context & Reaction

Anthropic set pricing for Fable 5 at $10 per million input tokens and $50 per million output tokens. The model is included on Pro, Max, Team, and seat-based enterprise plans through June 22. Starting June 23, the company said use will require credits unless capacity allows an extension.

Additionally, Anthropic launched Claude Mythos 5 for a smaller group of approved users, using the same underlying model as Fable 5. Mythos 5 starts through Project Glasswing in cooperation with the U.S. government, including cyberdefenders and critical software infrastructure providers. The company said Mythos 5 lifts some safeguards for approved cybersecurity users and plans a trusted access program for selected biology researchers and companies.

Anthropic said business customer traffic on Mythos-class models will face 30-day retention but will not use this data to train new Claude models. As of today’s announcement, roughly 95% of Fable sessions involve no fallback to Opus 4.8, though the safeguards trigger in less than 5% of sessions on average.

Background & Historical Context

The Fable 5 release marks the latest milestone in Anthropic’s development of increasingly capable AI models. The company said the cybersecurity controls cover exploitation and other offensive cyber tasks, while biology and chemistry safeguards address many requests due to dual-use risks.

The decision to route some requests to Claude Opus 4.8 reflects Anthropic’s conservative approach to releasing powerful AI systems. The fallback applies specifically to selected cybersecurity, biology, chemistry, and distillation-related queries, though the company acknowledged the safeguards may catch some harmless requests.

This measured release strategy aligns with broader industry trends as AI companies balance capability advancement with responsible deployment. The inclusion of Project Glasswing, involving U.S. government cooperation, suggests Anthropic is working closely with regulatory bodies on safety frameworks.

What This Means

The launch signals Anthropic’s commitment to pushing AI capabilities while implementing guardrails for high-risk applications. Users can expect:

– Short-term impact: Fable 5 availability through June 22 on existing plans gives developers and enterprises a window to test the model’s capabilities in software engineering, vision, and scientific research tasks. The conservative safety tuning means some legitimate queries may be misrouted initially, though Anthropic plans to reduce false positives.

– Long-term implications: The tiered access model for Mythos 5, starting with government-approved cyberdefenders and later biology researchers, suggests Anthropic may expand trusted access programs. This could set a precedent for how powerful AI models are distributed across sensitive domains.

– User action items: Developers should test applications against Fable 5’s API immediately to understand where fallbacks occur. Enterprise customers should evaluate their needs against the June 23 credit-based pricing model, while users in cybersecurity and biology fields may want to explore Mythos 5 access through the appropriate channels.

This is not financial advice. Conduct your own research before integrating these models into your workflow.

—