Claude Fable 5.2

On November 9, 2026 (Anthropic announcement), Anthropic released Claude Fable 5.2 for multi-hour agent sessions, large code migrations, and deep technical research. On the Remote Labor Index built from real freelance projects, it delivered 41.3% of jobs at least as well as a human (Fable 5.1: 17.9%, Opus 5.5: 21.3%, GPT-6 Astra: 20.8%). On the FrontierCode 2.0 agentic coding test it scored 49.1% at Xhigh effort, the highest of any released model tested. It generates output up to 30% faster than Fable 5.1, and API pricing dropped from $10 / $50 to $8 / $40 per million tokens (20% cheaper).

How to turn it on

• Claude (Web, Desktop, Mobile): available on all paid plans via the model selector.
• Claude Code: update to v2.1.321+ and select Fable:

claude update
claude --model fable

Inside an active session, run /model fable.

• Claude Platform API: model ID claude-fable-5-2 (1M context window, 128K max output).

Limits, usage credits, and effort settings

• Plan limits and usage credits: On Pro plans and standard Team seats, Fable can use up to 50% of your weekly usage limit. In Claude Code, depending on your plan and seat, Fable can bill to usage credits instead of your plan's limits: if the model picker shows "Requires usage credits" next to Fable, that's you, and headless claude -p runs bill credits without asking. Eligible paid subscribers get 1 to 3 bonus Fable resets this week, based on how long they've subscribed.
• Effort (effort) levels: The default is medium in Claude apps and Claude Code, where Fable 5.2 finishes hard tasks with fewer tokens than Fable 5.1. On FrontierCode 2.0 at high effort, it scored about level with Opus 5.5 at max effort for roughly half the cost per task (Anthropic announcement).
• When to pick Opus 5.5 vs Fable 5.2: Anthropic recommends starting daily coding and knowledge tasks with Opus 5.5, and stepping up to Fable 5.2 for open-ended migrations, multi-hour audits, or as an Orchestrator/Advisor directing smaller subagents.

3 ready prompts for long-horizon jobs

1. Multi-file codebase migration

Read this repository end to end and write an incremental migration plan for [GOAL, e.g. porting our Python service to Rust] where every step is independently testable. Save the plan to PLAN.md, then execute each step in order. Run the test suite after every step and fix any regressions before moving on. Finish with a summary of benchmark gains and untouched files.

2. Subagent security and edge-case audit

Spin up subagents to audit authentication, billing, permissions, and data-deletion flows across this codebase. Identify untested edge cases, write fixes and tests for the 10 highest-risk paths, and run the full test suite to verify. Do not report any finding you haven't verified in code.

3. Deep technical research and Architecture Decision Record (ADR)

Research [TOPIC / STACK DECISION] using official documentation and published benchmarks, citing the source and date next to every claim. Deliver a one-page Architecture Decision Record covering the options, cost trade-offs, failure modes, and your final recommendation.

Links

• Anthropic: Introducing Claude Fable 5.2 and Claude Mythos 5.2
• Claude Platform Docs: Claude Fable 5.2 Overview
• Claude Code: Model Configuration