Anthropic releases Claude Fable 5, a new AI model for creative work
Anthropic announced Claude Fable 5 on 10 June, the first publicly available model in its Mythos family. The model is designed for creative tasks - drafting stories, generating design concepts, producing documentation - alongside software engineering and image analysis. Pricing is set at roughly twice the cost of Claude Opus 4.8.
The model handles multi-stage problems and visual inputs more reliably than earlier Claude versions, according to Anthropic's benchmarks. It scores about 5% higher than Claude Opus 4.8 on overall performance measures and currently leads publicly available AI systems.
How safety guardrails work
Fable 5 applies strict restrictions in high-risk areas: cybersecurity, biology, chemistry, and model distillation. When a user requests a restricted action, the model blocks the task and falls back to Claude Opus 4.8 instead of attempting the work.
Anthropic tested these safeguards through an external bug bounty programme and red team testing. After more than 1,000 hours of attempts to bypass the protections, the company reported no universal jailbreak that worked consistently.
The company will also offer Mythos 5, a variant with fewer restrictions, to vetted groups through Project Glasswing.
Who gets access first
Fable 5 access is expanding through Project Glasswing to 150 organisations across 15+ countries, including India, as well as select partners from the Five Eyes alliance and NATO. Anthropic said it plans to widen eligibility over time.
What this means for creative teams
AI for Creatives typically aims to speed up ideation and drafting work. Fable 5's improved reasoning and visual understanding could help studios, marketers, and developers move faster on concept generation and code support.
The fallback to Opus 4.8 for restricted tasks sets a clear boundary. Some research uses may hit those limits, but the staged rollout through Glasswing clarifies where advanced capabilities appear first.
Your membership also unlocks:
