Anthropic releases Fable 5, bringing unreleased Mythos capabilities to public subscribers
Anthropic has launched Fable 5, a new AI model family derived from its previously restricted Mythos system. Available to Claude subscribers until June 22, the release marks a significant shift in the company’s strategy to deploy advanced capabilities while managing safety risks.

Anthropic has officially released Fable, a new family of artificial intelligence models designed to bring the capabilities of its unreleased Mythos system to the general public. The initial release, Fable 5, is available to Claude subscribers at no additional cost until 22 June, after which usage will require the purchase of credits. The model starts at version five to align with the company’s existing numbering conventions and represents a substantial leap in performance for the San Francisco-based firm.
According to Anthropic, Fable 5 outperforms its previous flagship, Opus 4.8, as well as competitor models GPT-5.5 and Gemini 3.1 Pro. The company noted that the model’s lead over other systems becomes more pronounced in longer and more complex tasks. Specific strengths include software engineering, drug design, and document analysis, areas where the firm has historically sought to improve its standing in the competitive AI market.
A notable improvement in Fable 5 is its vision capability, addressing weaknesses seen in earlier iterations such as Claude 3.7 Sonnet. While previous models struggled with pixel-based interfaces, requiring external overlays to interpret visual data, Fable 5 can play Pokémon FireRed using a minimal vision-only harness. The model is also capable of extracting precise data from scientific figures and rebuilding web application source code from screenshots alone.
To mitigate safety risks associated with the model’s increased power, Anthropic has implemented safeguards that route prompts on certain sensitive topics through the less-capable Opus 4.8 model. The company stated that these triggers occur in less than 5 percent of sessions on average. Anthropic acknowledged that the safeguards are tuned conservatively and may occasionally catch harmless requests, but the firm is working to reduce false positives as more capable models are released.
Pricing for usage credits after the free period is set at $10 per million input tokens and $50 per million output tokens. Anthropic also released Mythos 5, which shares the underlying architecture with Fable 5 but features fewer safeguards. This version will initially be available via Project Glasswing and a trusted access program. The firm intends to restore Fable 5 as a standard part of subscription plans once sufficient capacity allows.
The launch follows the April debut of the Mythos system through Project Glasswing, a program that shared access with select partners including Apple and NVIDIA to help harden software against AI cyberattacks. The initiative also influenced the White House to reconsider its policies on AI regulation. Anthropic has stated it aims to restore Fable 5 access as a standard subscription feature as quickly as possible.


