Fable 5 is the “safety-hardened” version of Mythos, bringing top-tier AI power to everyday users for the first time.
Anthropic officially launched Claude Fable 5 and Claude Mythos 5 on June 9, 2026, marking the first time the company has brought a Mythos-class model outside the scope of limited testing.
The story began in April 2026, when Anthropic introduced Claude Mythos Preview through Project Glasswing — a restricted access program available only to a small group of cybersecurity partners and infrastructure providers including Amazon Web Services, Apple, Google, Cisco, Microsoft, and JPMorgan Chase. The reason for keeping it under wraps was also the reason Anthropic spent two full months working to resolve: the model was too capable at finding and exploiting software vulnerabilities.
Fable 5 is the answer to that problem. At its core, it is the same model as Mythos, but equipped with an additional classifier system — specialized AI models running in parallel to detect and block dangerous requests before the main model processes them. When the classifier triggers, instead of a flat refusal, the system switches to Claude Opus 4.8 for processing and notifies the user. Anthropic says this mechanism activates in fewer than 5% of sessions, meaning the vast majority of users will receive an experience equivalent to Mythos 5.
Superior performance across multiple domains
On the SWE-Bench Pro coding benchmark, Fable 5 scores 80.3%, compared to 69.2% for Opus 4.8 and 58.6% for OpenAI’s GPT-5.5. In a real-world test, Stripe reported that Fable 5 completed a full migration of a 50-million-line Ruby codebase in a single day — a task that would take an entire engineering team more than two months to do manually.
For business analytics tasks, Fable 5 is the first model to surpass the 90% mark on Hex’s complex analytics benchmark — a full 10 percentage points ahead of Opus 4.8. IMC reported that the model scored “near perfect” on their trading analytics evaluation suite, which includes fact lookup, reasoning, root cause analysis, and expected value calculation.
On the visual side, Fable 5 can play through the entire Pokémon FireRed game using only screenshots, without any supporting tools — something previous Claude versions were unable to do even with additional assistive tools. The model also maintains better focus when working with millions of tokens across extended tasks.
Mythos 5 for licensed groups
Simultaneously, Claude Mythos 5 launched on the same day — this is the base model with full capabilities and no cybersecurity filters. Current access remains limited to Project Glasswing partners and a select group of biology researchers. Anthropic says it will gradually expand access through a trusted access program in the coming period.
In the biomedical field, Mythos 5 helped Anthropic’s in-house expert team accelerate certain drug design steps by approximately 10 times. In a study involving 14 protein targets, Mythos 5 independently planned, ran tools, and handled errors, generating promising drug design candidates for 9 of the 14 targets.
Pricing and access timeline
Both models share the same pricing: $10 per million input tokens and $50 per million output tokens — more than half the cost of the previous Mythos Preview. For API users and pay-as-you-go enterprise customers, Fable 5 is available immediately via the model ID “claude-fable-5”.
For Pro, Max, Team, and Enterprise subscription users, Fable 5 is included at no extra cost from now until June 22, 2026. From June 23, Anthropic will temporarily withdraw Fable 5 from these plans and shift to usage-based billing, until infrastructure capacity allows the model to return to fixed-plan inclusion.
This is also a notable moment as Anthropic has just filed a confidential IPO, placing the company on a path toward public markets alongside OpenAI and SpaceX.