Anthropic Opens the Vault: Claude Fable 5, Its Most Powerful AI, Is Now Public
Anthropic released Claude Fable 5 on June 9, making its first Mythos-class model publicly available to developers and consumers. The model scores 95% on SWE-bench Verified, ships with hard-coded safety guardrails in high-risk domains, and doubles the price of its predecessor — a calculated bet that peak capability and responsible deployment can coexist, released paradoxically just days after the company warned AI was 'getting too dangerous.'
Anthropic spent years insisting that its most capable models were simply too powerful to release broadly. On June 9, that calculus changed. The San Francisco AI safety company made Claude Fable 5 — its first Mythos-class model — available to the public, offering consumers, developers, and enterprises access to capabilities the company had previously restricted to a vetted inner circle.
The launch arrived in a charged moment. Just days earlier, Anthropic researchers had co-signed an open letter warning that frontier AI systems were “getting too dangerous” to deploy without unprecedented oversight. Critics were quick to note the apparent contradiction. The company’s implicit answer: Fable 5 ships with the most elaborate safety architecture it has ever built into a public model.
A New Tier of Performance
On the software engineering benchmark SWE-bench Verified, Fable 5 scores 95% — a figure that eclipses Claude Opus 4.8’s 88.6% and sets a new public high-water mark. In third-party evaluations, the gains were even more striking in applied settings.
Analytics platform Hex reported that Fable 5 was “the first model to reach 90% on our core analytics benchmark,” a proprietary test measuring real-world data analysis quality. Genspark, an AI search and workflow company, found it “beat every other model” across their internal suite, particularly on UI design generation and game coding. App-building platform Base44 highlighted its superiority in generating complex, multi-component applications from natural language prompts.
The model excels in the three domains Anthropic prioritizes: software engineering, knowledge work, and vision. For code-heavy enterprise workflows, the benchmark numbers carry direct business weight — companies paying for AI coding assistance are acutely aware of the difference between a 75% and a 95% success rate on real tasks.
Alongside Fable 5, Anthropic also released Claude Mythos 5, an updated version of the underlying frontier model, to organizations that had already been approved for Mythos access through Anthropic’s restricted program.
Safety by Architecture, Not Policy
What distinguishes Fable 5 from simply releasing an unconstrained frontier model is its hardcoded guardrail system. In four high-risk domains — cybersecurity, biology, chemistry, and nuclear or radiological material synthesis — Fable 5’s responses are blocked at the architecture level and the system falls back to Claude Opus 4.8, which answers with the model’s standard safety training.
This two-tier approach represents a significant engineering departure. Rather than relying on fine-tuning Fable 5 to refuse dangerous requests — a method that has proven porous across the industry — Anthropic built the restriction into the routing layer itself. The fallback is automatic and invisible to users.
The results appear robust. Anthropic ran an external bug bounty program involving over 1,000 hours of adversarial testing and reported zero universal jailbreaks. Early deployment data shows that approximately 95% of sessions run entirely on Fable 5 responses, meaning the fallback triggers rarely in practice — less than 5% of sessions.
“The model includes safeguards that would make this risk manageable,” Anthropic said in its release documentation, framing the release as a demonstration that safety and capability are not inherently at odds.
The Price of Power
Fable 5 comes with a significant cost increase. Pricing is set at $10 per million input tokens and $50 per million output tokens — exactly double what enterprises pay for Opus 4.8. For applications that make frequent or high-volume API calls, the economics require careful recalibration.
To soften the transition, Anthropic made Fable 5 available at no extra charge through June 22 to subscribers on Pro, Max, Team, and seat-based Enterprise plans. After that date, access shifts to a usage-credits model, with a return to standard plan inclusion at a later, unspecified date.
The model is also available on Amazon Bedrock, extending access to the large base of AWS enterprise customers who manage AI workloads through cloud infrastructure.
A Controversial Data Retention Change
Tucked into the release announcement was a policy change that drew sharp criticism from enterprise users: all Fable 5 and Mythos 5 API traffic is now subject to a mandatory 30-day retention period.
For developers who chose Anthropic’s API specifically for its zero-retention guarantee — a promise that requests and responses would never be stored — this marks a material breach of expectations. Anthropic framed the retention as a security necessity, explaining that logs are required to identify novel jailbreak attempts and coordinate responses to emerging attack patterns.
Privacy-focused legal and compliance teams at large enterprises are likely to treat this change as a renegotiation of risk. Several developer communities flagged the move as a reason to evaluate alternative providers.
The Paradox at the Center
The timing of Fable 5’s release is impossible to separate from its context. Anthropic’s researchers have positioned the company for years as the safety-first alternative in a race they believe could end in catastrophe. Releasing the most powerful public AI model in the world days after signing onto warnings about AI danger is, at minimum, a communications challenge.
The company’s answer — that robust safety architecture makes such releases responsible rather than reckless — is philosophically coherent. Whether it is practically adequate depends on whether the 5% of sessions that trigger guardrails are the right 5%, and whether the 30-day log retention creates new attack surfaces of its own.
What is not in dispute is the strategic logic. As OpenAI moves forward with its IPO and Google continues embedding Gemini across its ecosystem, Anthropic needed to demonstrate that its most capable models are not merely safe in a vault. Fable 5 is the company’s clearest answer to the question of whether responsible AI development can also be commercially competitive.
For the developers who build on it, the answer will arrive in the weeks ahead — measured in benchmarks, billing cycles, and the eventual scrutiny of whatever the first 5% of blocked sessions turn out to be.