Anthropic Drops Claude Fable 5 And Mythos 5 

• Fable 5 is now generally available on the Claude API and Amazon Bedrock. It includes built-in safety classifiers that automatically trigger fallbacks to safer models when users hit high-risk queries in areas like cybersecurity, biology, chemistry, or model distillation.

• Mythos 5 offers the full capabilities without those guardrails and is only available to approved partners.

• Both models feature a 1 million token context window and up to 128K output tokens 

• Early testing suggests Fable 5 is particularly strong at long-running, complex tasks — including large-scale coding projects, bug backlog clearance, and multi-step agentic work.

In a significant move for the AI arms race, Anthropic has officially brought its most advanced model class to the broader market with the launch of Claude Fable 5.

The company also released Claude Mythos 5, the unrestricted version of the same underlying model. However, Mythos 5 remains limited to vetted users through Project Glasswing (primarily for cyber defense and critical infrastructure work).

Anthropic described Fable 5 as delivering “state-of-the-art” results across software engineering, knowledge work, and vision benchmarks. The model is positioned as a step above previous Opus-class releases in sustained performance and attention to detail.

This release follows Anthropic’s earlier preview of Mythos-class models and comes after the company had been cautious about broadly releasing such high-capability systems. By creating two versions — one heavily safeguarded for general use and one less restricted for approved users — Anthropic is threading the needle between capability and safety concerns.

The move puts additional pressure on competitors. OpenAI and Google have been iterating on their own frontier models, but Fable 5’s combination of massive context, strong coding performance, and one-shot task execution sets a high bar for what developers will now expect from top-tier models.

Early user reports indicate the model is token-hungry and slower than previous generations, which aligns with the pattern seen in other recent frontier releases. It appears best suited for high-value, complex workflows rather than everyday chat or light tasks.

Anthropic has not disclosed exact pricing details yet beyond noting it will be more expensive than previous Opus models, consistent with the increased capability.

The launch gives developers and enterprises a new high-end option while keeping the rawest version of the technology in a more controlled environment. It will be interesting to see how quickly teams integrate Fable 5 into production agent workflows and whether the safety fallback system proves seamless in practice.

More details are available in Anthropic’s official announcement.