Skip to main content

Claude Fable 5.1 in Claude Code — what changes

Anthropic shipped Claude Fable 5.1 earlier this month. The model is now the default in Claude Code, and its changes are deep enough to justify a fresh read of modules 4 through 10 of this course. This module walks through what moves, module by module, and points at what you should re-read on your side.

What Fable 5.1 changes on the model side​

  • Longer context window — output token allocations have been raised; long plan write-ups and recaps no longer get cut mid-flow.
  • Chain of thought persisted across tools — when Claude Code chains search_web, then read_file, then edit_file, the reasoning thread is carried from one call to the next instead of restarting from scratch. In practice, that means fewer redundant tool calls and cleaner diffs.
  • Better handling of uncertain refusals — the model asks for confirmation when unsure about write permissions or scope, instead of refusing bluntly or half-executing. The permissions of module 7 become more readable as a result.
  • More stable error recovery — on a failing test, on a blocking lint, the model rolls back to the last valid checkpoint and proposes a targeted fix. The cycles described in module 8 (Plan mode and checkpoints) become genuinely reliable.

What this course integrates​

Every slash command, every CLAUDE.md example, every hook and every subagent in the course was re-run on Fable 5.1 between September 3 and September 9, 2026, on the Kiosque repository. The concrete edits:

  • Module 5 (quality and parallelism slash commands) was rounded out: /parallel now accepts an explicit token budget since Fable 5.1.
  • Module 8 (Plan mode and checkpoints) documents the new behavior of /plan --resume which now resumes exactly at the last auto-saved checkpoint, no longer at the start of the session.
  • Module 9 (hooks) notes that pre-tool hooks now receive the full reasoning thread; this lets you write safety guards that decide based on intent, not only on the command.
  • Module 15 (costs, context, team security) was re-measured: the average cost per turn dropped by 12 % on equivalent code-editing tasks.

What to re-read on your side​

If you took this course before September 3, 2026, three quick reads bring you up to date:

  1. Module 4 — updated table of session and configuration slash commands.
  2. Module 8 — new /plan --resume behavior with Fable 5.1 checkpoints.
  3. Module 9 — pre-tool hooks now receive the chain of thought: think through the security implications for your own hooks.

Three-line takeaway​

Claude Fable 5.1 makes Claude Code more reliable on long tasks, more predictable on failure/fix cycles, and a bit cheaper to use. This course was re-run on Kiosque with Fable 5.1 before publication: every command, every hook and every subagent you find here works as-is on the current version of the model.