Choose Your Claude Fable 5.2 Build Route
Run the loop yourself on the Messages API, meet it halfway with the Agent SDK, or hand runtime and tool execution to Anthropic-managed agents — each step trades control for less upkeep.
None

None

None

Long Story Video Skill

Turn your idea into 10s–10min video—AI writes scripts, prompts, and generates footage automatically.

Ads Video Skill

Ads Video Skill

Generate professional ads and sales videos—AI auto-generates scripts, prompts, and footage.

3D Science Explainer Video Skill

Convert scientific concepts into stunning 3D explain animations

AI Video Prompt Generator

Feedback

freeTrialImage.bannerPity

freeTrialImage.upgradeUnlock

  • ✓freeTrialImage.benefitHd
  • ✓freeTrialImage.benefitWatermark
  • ✓freeTrialImage.benefitUnlimited

claude fable

Meet Claude Fable 5.2: a 1M-token window plus effort-controlled adaptive thinking, engineered for hard reasoning and extended agentic work.

All Tools

Discover our comprehensive AI-powered animation toolkit

Where Claude Fable 5.2 Earns Its Place

Engineered as the next tier above Opus for demanding reasoning and extended agent loops, this model ships with the same spec sheet as Claude Mythos 5.2.

  • Made for Heavy Reasoning Loads
    If cranking effort on Claude Opus 5.5 still leaves your evals short, this is the tier to move to for complex reasoning and agents that run for a long time.
  • A Million Tokens In, 128k Out
    Default and maximum context both reach one million tokens across Claude Fable 5.2 and Claude Mythos 5.2, with responses of up to 128k tokens.
  • Thinking Depth Set by Effort
    There is no off switch for thinking. The effort parameter instead dials its depth through low, medium, high, xhigh, and max, and the API defaults to high.

Getting Claude Fable 5.2 Into Production

Settle on a build route, justify the upgrade with eval numbers, and resolve every breaking change before real traffic arrives.

What Ships Inside Claude Fable 5.2

On-demand compaction, swapping tools mid-conversation, a dedicated computer use toolset, fast mode, strict tool use, and structured refusal handling.

On-Demand Compaction (Beta)

With the compact-2026-09-04 beta header you choose exactly when earlier turns are summarized, keeping history append-only and the prefix check happy.

Swap Tools Mid-Conversation (Beta)

Rather than rewriting the tools array, add or drop tools inside a system message; send the inline-tools-2026-09-15 header on the Claude API, Claude Platform on AWS, and Anthropic-managed Google Cloud.

Computer Use Through a Toolset

Access computer use through computer_toolset_20260801; declaring the older computer_20251124 tool now returns a 400 invalid_request_error everywhere but Amazon Bedrock.

Where Fast Mode Runs

Fast mode is available on the Claude API, on Anthropic- and Google-managed Google Cloud, and inside Claude in Microsoft Foundry as a separate model, claude-fable-5-2-fast.

Strict Tool Use for Valid Schemas

To get schema-valid JSON, keep tool_choice on auto and set strict: true or move to JSON outputs; forcing a tool returns an error, and customer-managed encryption keys exclude structured outputs.

Refusals, Fallbacks, and Billing

Refusals arrive as HTTP 200 with stop_reason refusal plus a stop_details object; some categories may retry on Claude Opus 4.8 or Claude Opus 5, and fallback credit covers the prompt-cache cost.

FAQ

Claude Fable 5.2: Your Questions Answered

Straight answers on getting started, migrating safely, and what Claude Fable 5.2 costs.

1

Do I begin with Claude Opus 5.5 or Claude Fable 5.2?

For nearly every workload, begin on Claude Opus 5.5. Move to Claude Fable 5.2 only after Opus at higher effort still misses your bar.

2

What causes a 400 when I edit an earlier turn?

The prefix check compares the system prompt, tools, and messages ahead of a thinking block — change any of them and the call fails. A temporary workaround is the thinking-binding-controls-2026-08-01 header with prefix_mismatch_behavior set to drop_block.

3

Who can read Claude Fable 5.2 thinking blocks?

Just Claude Fable 5.2 and Claude Mythos 5.2 — they read one another's blocks. Blocks the active model cannot parse are stripped before delivery, and dropped blocks are not billed.

4

Is there a way to disable thinking or force a tool?

Not currently. Passing thinking enabled with budget_tokens, or thinking disabled, returns 400, as does tool_choice set to any or to a named tool. Use effort levels and strict tool use instead.

5

How does Claude Fable 5.2 pricing stack up against 5.1?

Input and output both cost less than Claude Fable 5.1, which charged $10 per million input and $50 per million output tokens. Cache reads remain at 0.025x base input, and batch jobs run $4 per million input tokens.

6

What does a refusal actually look like?

You get HTTP 200 carrying stop_reason refusal and a stop_details object that names the policy area. Certain categories can retry through server-side fallback, and charges vary by category.

Try Claude Fable 5.2 Before You Commit

Validate the upgrade first: confirm Claude Opus 5.5 no longer meets your bar, lock in a build route, then rehearse all four breaking changes on your own integration before real traffic shifts over.