Fable 5 is Anthropic's most capable publicly available model, leading in software engineering, scientific research, knowledge work, and vision. It's live today on all paid plans, in Claude Code, on the API, and across major cloud platforms.
The official getting-started thread for Fable 5 covers how to access it in Claude Code and on the Claude Platform, with model-switching instructions and usage guidance to follow.
Fable 5 is available in Claude Code starting today.
Fable 5 and Mythos 5 share the same underlying model. Mythos 5 has certain safeguards lifted and remains exclusive to Project Glasswing.
Switch to Fable in the Messages API or Claude Managed Agents by setting the model ID to "claude-fable-5". Some accounts may need an admin to enable access first. The /claude-api skill in Claude Code can help you build against it directly.
Cybersecurity and biosecurity requests can automatically reroute to Opus 4.8, with the model shown in the UI and billed at Opus prices. A new docs page and prompting guide cover the details.
Existing prompts and skills tuned for previous Claude models tend to be too prescriptive for Fable. The team recommends reviewing and potentially removing older instructions if the model's default behavior already beats them.
The recommended onboarding approach is to give Fable something harder than you'd expect prior Claude versions to handle. The suggestion: pick a week-long backlog item, let Fable interview you for the spec, enable auto mode, and check back in the morning.
Feedback loops with Fable work the same as with prior models: give it success criteria to check results against, via /goal in Claude Code or the Outcomes setting in Claude Managed Agents.
Run /model and pick Fable from the model picker, or go directly with /model claude-fable-5.
Agents can now spawn other agents, up to five levels deep. The depth=5 cap is a cautious starting point, and the feature ships in today's release. The framing is explicitly about managing context more effectively as tasks grow.
Multi-agent orchestration in Claude Managed Agents lets Fable delegate subtasks to dedicated agents running on smaller models, enabling more efficient and cost-conscious workflows.
An experimental approach: adding "fork:true" to a skill's frontmatter runs it in its own context window, and pairing that with agent-per-step instructions inside the skill keeps context clean across each step. Being tested on the built-in /code-review skill to improve performance.
