Full Breakdown
Anthropic Launches Claude Fable 5, First Public Mythos-Class AI Model with Safety Guardrails
6/9/2026, 11:56:45 PM
Public Release of Claude Fable 5
On 9 June 2026 Anthropic made Claude Fable 5 available to enterprise customers via the Claude API and its subscription plans. The model is positioned as the most capable AI system the company has ever released for general use. Pricing is set at $10 per million input tokens and $50 per million output tokens. When a request falls into a high-risk category (cybersecurity, biology, chemistry, or model distillation) the system automatically redirects the query to Claude Opus 4.8, the previous flagship model. Anthropic reports that such redirections occur in fewer than 5 % of sessions, and that 95 % of sessions run entirely on Fable 5 responses.
From Mythos Preview to Public Access
Anthropic first unveiled the Mythos class in April 2026, limiting it to a small cohort of partners under the Project Glasswing program because of the model’s advanced cybersecurity capabilities. The company warned that unrestricted release could enable malicious actors to automate sophisticated attacks. After extensive internal and external red-team testing, Anthropic announced the public-facing version (Fable 5) with built-in safeguards, while retaining an unrestricted version (Claude Mythos 5) for vetted cyber-defenders and critical-infrastructure providers.
Performance, Pricing, and Token Economics
Independent benchmarks show strong results: Hex recorded a 90 % score on its core analytics benchmark, the first model to exceed that threshold; SWE-Bench Pro yielded 80.3 %; and Hex noted a 10-point improvement over Opus 4.8. Analytics firm Hex and agentic platforms Base44, Genspark, and Rakuten all reported superior tool-calling, UI design, and “one-shot” app generation. The token cost is roughly double that of Opus 4.8 but less than half the price of the earlier Mythos Preview offering.
Safeguards, Data Retention, and Model Access
Anthropic states the safeguards are “conservatively tuned” to block potentially dangerous queries while informing users when a fallback occurs. The company also introduced a 30-day retention policy for business-customer traffic, asserting the data will be used only for security monitoring and not for training future models. Access to Claude Mythos 5 remains limited to Project Glasswing participants, with plans to expand through a “trusted-access program” in collaboration with the U.S. government.
Industry Adoption and Early Benchmarks
Stripe reported that Fable 5 migrated a 50-million-line Ruby codebase in a single day, a task that would normally require two months of engineering effort. Hex highlighted the model’s strong judgment on complex analytical tasks, while Genspark found it outperformed all tested competitors on UI-design and game-coding challenges. Rakuten emphasized the model’s ability to self-validate its work, describing it as essential for highly autonomous operations.
Criticisms, Cost Concerns, and Guardrail Limits
Security analysts note that the $10 / $50 token rates may deter smaller firms, especially as AI budgets tighten. Some observers argue the guardrails could “trip at the faintest hint of a security problem,” potentially restricting legitimate defensive research. The reliance on fallback to Opus 4.8 also raises questions about consistency of output quality across tasks.
Conflicting Reports and Unresolved Details
Sources differ on the output-token price: one report lists $25 per million output tokens, while the majority cite $50. Reported partner counts for Project Glasswing vary between roughly 50 organizations (initial cohort) and 150 organizations after the recent expansion. Data-retention policies are described as 30 days in some statements, but other sources omit this detail, leaving the full scope unclear.
Timeline of Key Milestones
- April 2026: Claude Mythos Preview launched to ~50 partners under Project Glasswing.
- 9 June 2026: Public launch of Claude Fable 5; simultaneous release of Claude Mythos 5 to vetted partners.
- 22 June 2026: End of free inclusion of Fable 5 in Pro, Max, Team, and Enterprise plans.
- 23 June 2026 onward: Usage credits required for Fable 5; plans to restore standard access when capacity permits.
Verbatim Quotes
- “In a press release, the company wrote: “Fable 5’s capabilities exceed those of any model we’ve ever made generally available.” — Anthropic spokesperson
- “Internally, we ran an external bug bounty that produced no universal jailbreaks in over 1,000 hours of testing. We then worked with external red-teaming orgs which also failed to find universal jailbreaks.” — Anthropic security team
- “Fable 5 is a real step forward for the developers GitHub serves,” — GitHub spokesperson
- “In its blog announcement, Anthropic wrote: “The capabilities of models like Fable 5 and Mythos 5 have the potential to do profound good for the world.” — Anthropic blog
- “3% on SWE-Bench Pro, a rigorous software engineering benchmark, and became the first AI model to cross the 90% threshold on Hex's analytical benchmark.” — WION report
Outlook and Expansion Plans
Anthropic plans to broaden the trusted-access program, inviting additional cybersecurity and life-science organizations while maintaining the current guardrails. The company also indicated that future frontier models will incorporate similar safety mechanisms and data-retention policies. The public release of Claude Fable 5 therefore serves as a test case for scaling powerful AI capabilities responsibly across the enterprise sector.
