Anthropic Releases Claude Fable 5: Its Most Powerful Model Ever Made Publicly Available, Built on the Mythos Architecture
Anthropic launched Claude Fable 5 on June 9, 2026, making publicly available for the first time a model built on its Mythos‑class architecture, a capability tier the company had previously considered too powerful for general release. The announcement came alongside the simultaneous release of Claude Mythos 5, a restricted version of the same model that retains capabilities Fable 5 has been safeguarded against. Both models are available immediately through the Claude API, Claude.ai, the Claude mobile apps, Claude Code, and consumption‑based Enterprise plans. Availability on subscription plans including Pro, Max, and Team tiers begins today through June 22 at no additional cost, after which Anthropic says demand patterns will guide how access is managed going forward.
From Restricted Preview to General Availability
Claude Mythos Preview was first released in April 2026 through Project Glasswing, a restricted programme involving a small number of trusted partners including AWS, Microsoft, Apple, and CrowdStrike. The rationale for the limited release was explicit: the model demonstrated an exceptional ability to autonomously discover and chain zero‑day exploits across major operating systems and browsers, capabilities that Anthropic said made broad public access too risky without additional safeguards.
Claude Fable 5 is Anthropic's answer to that constraint. It shares the same underlying weights as Mythos 5, meaning the intelligence and reasoning architecture are identical. The difference is a safeguard layer built around classifiers that identify queries touching cybersecurity, biology, chemistry, and model distillation. When those classifiers fire, the request is automatically routed to Claude Opus 4.8 rather than answered by the Mythos‑class system. Anthropic says this fallback affects fewer than five percent of sessions, and critically, users are not charged Fable 5 prices for responses handled by Opus.
The naming is deliberate. Fable comes from the Latin fabula, meaning that which is told, cognate with the Greek mythos. Anthropic positioned Mythos‑class models as a tier above its Opus line, and Fable 5 is the first member of that tier available to general users.
What the Benchmarks Show
The performance gap between Fable 5 and its predecessors is substantial rather than incremental. On SWE‑bench Verified, which evaluates a model's ability to resolve real GitHub issues in open‑source repositories, Fable 5 scores 95.0 percent. Claude Opus 4.8 scores 88.6 percent on the same benchmark. On SWE‑bench Pro, which tests harder, longer‑horizon software engineering tasks, Fable 5 posts 80.3 percent, eleven points above Opus 4.8's 69.2 percent and more than twenty points ahead of GPT‑5.5 at 58.6 percent. Anthropic also reports that Fable 5 reaches 80.4 percent on FrontierCode Diamond at maximum effort, a benchmark for difficult production‑codebase‑standard tasks where the company says it leads all available models.
Beyond coding, the model shows strong performance on GPQA Diamond, a test of reasoning in advanced technical domains including physics, chemistry, biology, and mathematics, where it scores 94.6 percent.
The context window is one million input tokens with 128,000 output tokens. Extended thinking is supported. The model handles both text and vision inputs.
Andrej Karpathy, a prominent AI researcher and former Tesla and OpenAI engineer, described Fable 5 as a major‑version‑bump‑deserving step change forward. That characterisation is consistent with the benchmark gaps, which are among the largest single‑release improvements the public model ecosystem has seen in recent months.
Capabilities and What They Mean in Practice
Anthropic describes Fable 5 as its new generation of intelligence for the hardest problems in knowledge work and software engineering. The autonomous operation aspect is among the most practically significant changes. The company says both Fable 5 and Mythos 5 can work autonomously for longer than any previous Claude model, meaning they are capable of completing multi‑step tasks over extended periods with less human intervention, an improvement that directly addresses the reliability limitations that have constrained earlier agentic AI deployments.
The engineering implications are concrete. Anthropic cited, among other examples, an instance in which Fable 5 completed a migration across a 50‑million‑line codebase in a single day, a task that would typically require substantial engineering time to coordinate and execute.
Memory and long‑context handling have also been improved, supporting the kinds of extended‑session, document‑heavy workflows that enterprise customers in legal, finance, and research are increasingly trying to automate.
Pricing and Access
Claude Fable 5 is priced at $10 per million input tokens and $50 per million output tokens, with prompt caching available at a 90 percent discount. This is double the price of Opus 4.8 but less than half the cost of Claude Mythos Preview, making Mythos‑class capabilities accessible to a substantially wider range of developers and enterprises than was previously possible.
The model is accessible on the API under the identifier claude‑fable‑5. It is available through AWS Bedrock, Google Cloud Vertex AI, and Microsoft Azure AI Foundry. Anthropic has flagged that demand is expected to be very high and difficult to predict, and has warned that subscription access may be managed in stages depending on capacity.
Claude Mythos 5, the unsafeguarded configuration of the same model, remains restricted to vetted partners through Project Glasswing for cybersecurity‑specific applications, continuing the framework established with the original Mythos Preview release.
For full documentation, benchmarks, and access information, visit the official Anthropic model page at https://www.anthropic.com/news/claude‑fable‑5‑mythos‑5.
The release of Fable 5 marks the moment Anthropic moved from offering its frontier capability to a handful of partners to making it broadly available to the developer and enterprise market. With an IPO that could value the company close to a trillion dollars expected later this year, the commercial timing of this launch is not incidental. For developers and enterprises that have been waiting for a public model with Mythos‑class capabilities, that wait ended today.
