.png?w=1024)
Anthropic is finally letting everyday users take its most formidable AI architecture for a spin.
The San Francisco-based AI startup on Tuesday launched Claude Fable 5, a publicly available version of its vaunted Mythos-class model.
Originally unveiled in April, the core Mythos technology was famously locked away from the mainstream due to acute fears over its potent hacking and biological capabilities. By implementing a strict fallback system, Anthropic believes it has safely split the difference between commercial demand and national security.
Splitting the difference on safety
Fable 5 is built on the same underlying architecture as its restricted sibling, but it comes equipped with specialized "safety classifiers." These separate AI systems act as gatekeepers, scanning incoming prompts for high-risk topics.
If a user tries to probe Fable 5 about offensive cybersecurity tactics, chemistry, biology, or model distillation, the system automatically redirects the query. Instead of giving a flat refusal, the platform routes the task to a lower-tier model, Claude Opus 4.8, which was released in late May.
Anthropic reports that this fallback triggers in fewer than 5% of user sessions on average, leaving the remaining 95% of interactions to run at full Mythos-level strength.
To prove the robustness of these barriers, the company subjected Fable 5 to over 1,000 hours of external "red-teaming" and ran a dedicated bug bounty program. According to Anthropic, the gauntlet yielded zero successful "universal jailbreaks" capable of completely unlocking the model.
Breaking benchmark records
When the safeguards aren't triggered, Fable 5 demonstrates a massive leap forward in long-form, complex engineering and analytical tasks.
Early testing partners reported immediate efficiency spikes. Stripe noted that Fable 5 successfully migrated a massive 50-million-line Ruby codebase in a single day, a project they estimated would have taken an entire engineering team over two months to complete manually.
Independent benchmarks paint a similar picture of dominance:
- SWE-bench-pro: Fable 5 achieved a top score of 80.3% in agentic coding skills, comfortably outpacing OpenAI's GPT-5.5 (58.6%) and Google DeepMind's Gemini 3.1 Pro (54.2%).
- Hex finance benchmark: The analytics company Hex revealed that Fable 5 is the first model to break the 90% threshold on its core benchmark for complex, long-running senior-level reasoning tasks.
- Vision and autonomy: In stress tests, the model proved capable of rebuilding web applications directly from screenshots. It even successfully beat Pokémon FireRed using a bare-bones, vision-only harness, a feat that had repeatedly stumped previous iterations of Claude.
The unlocked tier: Claude Mythos 5
Simultaneously, Anthropic launched Claude Mythos 5, the fully unrestricted version of the model, tailored explicitly for the defense sector.
Mythos 5 is currently being deployed via "Project Glasswing," a joint cybersecurity initiative run in collaboration with other AI leaders. The project's trusted partner pool was recently expanded in early June to encompass roughly 200 organizations across more than 15 countries.
Beyond defense, Anthropic's internal scientists have leveraged the unrestricted Mythos 5 to accelerate protein design pipelines by roughly 10x, autonomously executing intricate workflows such as selecting binding sites and recovering from software errors without human intervention.
The company plans to introduce a "trusted access program" soon to gradually extend these biology and cyber-defense capabilities to verified academic and corporate researchers.
High intelligence, higher price tag
The bleeding edge of AI reasoning doesn't come cheap. Both Fable 5 and Mythos 5 are priced at $10 per million input tokens and $50 per million output tokens, making the lineup twice as expensive to run as Claude Opus 4.8.
The premium pricing reflects the steep infrastructure costs weighing on the startup. Anthropic remains far from profitable and is currently burning through capital to secure computing power, including a massive $1.25 billion-per-month data center lease from Elon Musk’s xAI.
Anthropic recently closed a $65 billion Series H funding round, pushing its private valuation to a staggering $965 billion and pushing it past rival OpenAI ($852 billion).
With both AI heavyweights having confidentially filed their IPO prospectuses with the SEC over the past week, Fable 5's commercial rollout will serve as a critical test of whether Anthropic can convert its technical milestones into sustainable public market value.
Also read: Anthropic warns that AI systems may soon help build their own successors, raising new questions for governance and enterprise risk.