Claude Code with Claude Fable 5.2: setup, cost and when to switch from Opus 5.5

Published 2026-11-09 · Last updated 2026-11-09

Short answer

To use Claude Fable 5.2 in Claude Code, run /model fable or start with claude --model fable. It costs $8 input and $40 output per million tokens, twice Opus 5.5. Keep Opus 5.5 as your default and switch for long, hard agent runs.

Pick Claude Fable 5.2 in Claude Code

Anthropic released Claude Fable 5.2 on November 9, 2026 (announcement). Claude Code needs v2.1.321 or later to run it. Run claude update if a request fails on an older version (Claude Code model docs). No plan uses a Fable model by default, so you pick it yourself. There are three ways.

1. The /model command

/model fable

/model with no argument opens a picker. Passing a name switches the model and saves it as your default for new sessions. The fable alias resolves to Fable 5.2, except in Claude apps gateway sessions, where you run /model claude-fable-5-2 instead.

2. The --model flag

claude --model fable
# or the full model ID
claude --model claude-fable-5-2

--model sets the model for that session only. It takes an alias or a full model name and overrides both the model setting and the ANTHROPIC_MODEL environment variable (CLI reference).

3. The model field in settings.json

{
  "model": "fable"
}

Put this in ~/.claude/settings.json for every project, or in .claude/settings.json to share it with your team (settings docs). If a project or managed settings file pins claude-fable-5-1, Claude Code keeps 5.1 there. Change it to fable or claude-fable-5-2.

The availableModels setting limits which models people can pick, for example to keep Fable out of a shared team config.

Running Claude Code through Amazon Bedrock? The models overview lists the Bedrock model ID anthropic.claude-fable-5-2 (models overview). On providers other than the Anthropic API, Claude Code's docs say to use the provider's model ID or pin it with ANTHROPIC_DEFAULT_FABLE_MODEL (model docs).

What Claude Fable 5.2 costs

Prices per million tokens, from Anthropic's models overview, announcement and Claude Code fast mode docs:

Fable 5.2 is 20% cheaper than Fable 5.1 and still twice the price of Opus 5.5 per token. Cache reads bill at 2.5% of the base input price, which works out to $0.20 per million tokens on Fable 5.2. Full rate card: pricing page.

Fast mode

Fast mode runs the same model up to 2.5x faster at a higher price per token. Turn it on with /fast, or set "fastMode": true in settings. Claude Code supports it on Fable 5.2, Opus 5.5, Opus 5 and Opus 4.8, not on Sonnet or Haiku (fast mode docs).

On Fable 5.2, fast mode costs $16 / $80, double the standard rate. On a Pro, Max, Ultra, Team or Enterprise plan, fast mode bills to usage credits, not your included limits. In Claude Code it works through the Anthropic API or a Claude subscription only, not through Bedrock, Google Cloud, Microsoft Foundry or Claude Platform on AWS. Use it when you are waiting on the answer. Leave it off for long runs you check on later.

Effort levels

Fable models in Claude Code support five effort levels: low, medium, high, xhigh and max. Higher effort means more thinking per step, which means more output tokens. In Claude Code, Fable 5.2 starts at medium unless you set a level (model docs). On the API the default is high (models overview).

/effort high            # in a session
claude --effort xhigh   # for one launch
CLAUDE_CODE_EFFORT_LEVEL=high claude   # overrides both

Anthropic's announcement gives one data point worth acting on. On FrontierCode 2.0, Fable 5.2 at high effort scored 45.3%, about level with Opus 5.5's best of 45.5%, at about half the cost per task. Those are Anthropic's numbers. Try high before xhigh and compare the result on your own repo.

Cost per session: the math

Claude Code bills per token, so a session's cost is input tokens times the input price plus output tokens times the output price. Run /cost (an alias for /usage) to see Claude Code's estimate for the current session. It computes that number locally at list price (costs docs).

Worked example. Take a long agent session that sends 1.5 million input tokens across all turns and gets back 120,000 output tokens, with no caching:

Three things change the real number:

• Caching. A long session rereads the same files every turn. If 1.35 million of those 1.5 million input tokens were cache reads at $0.20 per million, Fable 5.2's input cost drops from $12.00 to $1.47 ($1.20 + $0.27), before any cache-write charges.
• Token count is not fixed across models. A stronger model can finish in fewer turns. That is how Anthropic gets "half the cost per task" on FrontierCode despite the 2x token price.
• Opus 5.5 in fast mode costs the same per token as Fable 5.2 at standard speed ($8 / $40). If you turned on fast mode for Opus to save time, Fable at standard speed costs the same.

On a Pro, Max, Team or Enterprise plan you pay in usage limits, not dollars. Depending on your plan and seat, Fable usage can bill to usage credits instead of your included limits. The /model picker then shows "Requires usage credits" on the Fable row (model docs). Anthropic's launch offers article covers the subscriber terms for launch week.

When Opus 5.5 is enough

Anthropic's own advice is to start with Opus 5.5 for most workloads and use Fable 5.2 for demanding reasoning, long-horizon agentic work, or when Opus 5.5 at higher effort still falls short (models overview). The Claude Code docs describe Fable models as "suited to tasks larger than a single sitting." Opus 5.5 is also the default model on Pro, Max, Team, Enterprise and the API.

Our rule of thumb, based on that guidance:

• Stay on Opus 5.5 for single-file fixes, tests, code review, docs and anything you finish in one sitting.
• Switch to Fable 5.2 for multi-hour agent runs, refactors across many files, planning work that fans out to many subagents, and bugs Opus 5.5 failed to fix at xhigh.
• Switch back once the hard part is done. /model opus takes one command.

Benchmark rows from Anthropic's announcement (Anthropic's numbers, not independent tests):

Fable 5.2 does not top every row. Gemini 4 Carbon leads AutomationBench at 53.9%.

Independent numbers: Artificial Analysis lists Claude Fable 5.2 at max effort with an Intelligence Index of 65, ranked #1 of 74 models, as of 2026-11-09. The same page measures 101 output tokens per second and 171.16 seconds to the first answer token on Anthropic's API. Max effort thinks for a long time before it answers, which is one more reason to start at high.

Skills that pair well with Fable 5.2

From the AY CLAUDE catalog:

• writing-plans: Write the plan on Fable 5.2, then hand the steps to a cheaper model.
• subagent-driven-development: Runs one fresh subagent per task with review in between, which suits long Fable sessions.
• dispatching-parallel-agents: Splits independent failures across parallel agents instead of one long, expensive thread.
• create-subagents: Set up subagents so routine work runs on a cheaper model than your main session.
• claude-code-analyzer: Reads your Claude Code usage patterns so you can see where the expensive turns go.

Sources

• Introducing Claude Fable 5.2 and Claude Mythos 5.2 (Anthropic)
• Models overview and pricing (Claude Platform docs)
• Model configuration, fast mode, settings, CLI reference and costs (Claude Code docs)
• Claude Fable 5.2 launch offers (Claude support)
• Claude Fable 5.2 (max) (Artificial Analysis, checked 2026-11-09)