Skip to main content

anthropic/claude-fable-5.2/chat

Anthropic's Claude Fable 5.2 API brings 1,000,000-token context, 128,000-token output, five adaptive thinking effort tiers, and lower token pricing than Fable 5.

Input $7.20Output $36per 1 million tokens10% below the publisher's list priceCached input $0.72Cache write $9Anthropic official priceInput $8Output $40

Workbench

Claude Fable 5.2

What can Claude Fable 5.2 help with?

Missing the model you need?

FRONTIER AGENTIC & CODING API

Claude Fable 5.2 API at a glance

Anthropic's upgraded long-horizon reasoning and multi-repository engineering model at reduced token cost.

Context window
1M
Up to one million tokens across long codebases and documents.
Max output
64K
Long structured responses in a single turn.
Reasoning effort
4 TIERS
Switch between low, medium, high, and max.

Results

What Claude Fable 5.2 is good at

  • Adaptive depth

    Four reasoning tiers on a single endpoint

    Scale deliberation smoothly across low, medium, high, and max through one parameter instead of maintaining separate client integrations. Routine formatting stays crisp on low while intricate system audits tap max without altering your request format.

  • Four-tier billing

    Prompt caching engineered for repeat reads

    Pin large repositories, legal contracts, or manuscript bibles at the front of messages and settle repeat turns at $0.72/1M cached input tokens—90% below standard input and 10% below Anthropic list pricing across all four token tiers.

  • Multimodal window

    One million tokens across text and vision

    Combine long codebases, conversation transcripts, and image_url captures inside a 1,000,000-token budget, with up to 64,000 tokens of structured analysis or prose returned in a single pass.

  • OpenAI-compatible

    Drop-in SSE streaming and 128-tool control

    Point existing OpenAI SDK code to /v1/chat/completions with model ID anthropic/claude-fable-5.2/chat for live SSE streaming, named tool_choice enforcement, custom stop sequences, and sequential or parallel tool calls.

Before you run it

What to know before you run Claude Fable 5.2

Chat
Accepts text and image_url parts; audio, video, and binary attachments must be converted before submission.
Chat
Adaptive thinking manages sampling internally, so temperature and top_p are omitted from this endpoint.

Use cases

What to build with the Claude Fable 5.2 API

  • Cross-repository refactoring and migrations

    Load database schemas, API contracts, and integration test suites together inside the 1,000,000-token context window to draft synchronized migration steps and rollback checks.

  • Atmospheric long-form fiction and dialogue

    Keep an entire novel draft and character style bible pinned in context to write slow-burn romantic or noir continuations with consistent voice, sensory lighting, and subtext.

  • Visual UI and diagram inspection

    Pair screenshots, wireframes, or system topology images with source files via image_url blocks so Claude Fable 5.2 pinpoints layout regressions and state mismatches in one turn.

  • Stateful multi-tool engineering agents

    Register up to 128 function definitions and toggle parallel_tool_calls to false whenever sequential database checks or deployment steps must wait for prior tool outputs.

  • Deep root-cause debugging

    Switch reasoning_effort to high or max when untangling distributed race conditions, duplicate billing retries, or intermittent production deadlocks across multi-service logs.

Prompts

Claude Fable 5.2 prompt examples you can copy

  • Chat

    Candlelit Parisian study dialogue continuation

    Continue the scene below for 350 to 450 words. Keep the candlelit Parisian study atmosphere, the slow-burn tension, and the rhythm of whispered dialogue without rushing the beat. Rain ticked steadily against the tall balcony panes of the sixth-arrondissement study. Past midnight, the room held two small pools of light from beeswax tapers guttering beside a brass inkwell and scattered manuscript pages. Julien, thirty-six, sat back in his leather desk chair, his ivory silk shirt unbuttoned at the throat and sleeves rolled neatly to his forearms, a fountain pen resting across his knuckles. Camille, thirty-four, crossed the parquet floor barefoot, her emerald silk evening dress catching the warm flicker as she leaned close to set a crystal coupe beside his draft. The thin left strap slipped an inch down her bare shoulder; neither of them made a move to adjust it. She held his gaze across the narrowing space, her voice dropping to a low murmur. "You told the gallery you burned my October letter," she said softly, fingertips grazing the edge of the locked top drawer. "Yet the cedar scent is still on your collar."

    Why it works. Specifies character ages, wardrobe, candlelight, and word target so Claude Fable 5.2 sustains slow-burn romantic tension and crisp dialogue rhythm with balanced reasoning depth.

    Try this prompt
  • Chat

    Multi-service payment ledger migration review

    We are migrating a live immutable PostgreSQL billing ledger (180M rows) to monthly range partitions while introducing a transactional outbox worker that publishes settled-usage events to Kafka. Current constraints: - Peak write throughput is 1,400 ledger inserts/sec; read queries filter by (account_id, created_at DESC). - Settlement API workers hold short SERIALIZABLE transactions that verify account balance before inserting debit rows. - The new outbox poller runs 6 replicas with SELECT ... FOR UPDATE SKIP LOCKED in batches of 500. Audit this plan for: (1) lock contention or table-rewrite risks during partition cutover, (2) idempotency and ordering hazards if Kafka consumers process replayed outbox batches, and (3) a four-stage zero-downtime migration sequence with measurable abort thresholds.

    Why it works. Pairs concrete architectural artifacts with three named audit dimensions, making deep multi-step reasoning productive without wasting output on generic boilerplate.

    Try this prompt
  • Chat

    Distributed webhook retry race-condition plan

    Production incident: roughly 0.04% of payment top-up webhooks deposit funds into an account twice even though `wallet_ledger` has `UNIQUE (webhook_event_id)`. Observed handler logic (Go + PostgreSQL READ COMMITTED): 1. Handler begins transaction T1. 2. Calls `SELECT id FROM wallet_ledger WHERE webhook_event_id = $1` (returns 0 rows). 3. Calls external risk service over HTTP (takes 180-900ms while holding T1 open). 4. Calls `UPDATE accounts SET balance_cents = balance_cents + $1 WHERE id = $2`. 5. Inserts into `wallet_ledger` with `ON CONFLICT (webhook_event_id) DO NOTHING` and commits. Notice step 5 ignores conflict errors rather than rolling back step 4. Trace the exact interleaving between two concurrent webhook deliveries that produces the duplicate deposit, provide the minimal SQL and Go fix that eliminates both the duplicate deposit and the long transaction hold, and design a deterministic Go test using a synchronization barrier to reproduce and prove the fix.

    Why it works. Uses maximum reasoning depth for a narrow, high-stakes concurrency bug where tracing exact lock and isolation interleavings pays off.

    Try this prompt
  • Chat

    Midnight Seine hotel suite subtext rewrite

    Rewrite this two-page confrontation between two estranged auction-house rivals in their late thirties sharing a rain-soaked hotel suite overlooking the Seine at 1:00 a.m. Keep her burgundy satin robe loosely tied at the waist and his loosened black tie and open collar catching amber bedside lamplight. Trim overt exposition by 40%, deepen the murmured pauses, and end on a quiet line that leaves the reader guessing whether she reaches for the keycard or his collar.

    Why it works. Demonstrates targeted line-editing with measurable cuts and atmospheric wardrobe cues, keeping mature romantic tension refined and grounded.

    Try this prompt
  • Chat

    Cached manuscript continuity audit

    Using the static 60,000-word manuscript and character dossier pinned in the system prefix, audit Chapter 14's midnight conservatory meeting between Elena (35) and Marc (38). List every contradiction in wardrobe, candlelit room layout, earlier whispered promises, and timeline markers with exact chapter and paragraph citations.

    Why it works. Structures static reference material at the prefix so follow-up editorial passes hit the $0.72/1M cached input rate while requesting verifiable citations.

    Try this prompt
  • Chat

    Screenshot and CSS layout regression diagnosis

    Inspect the attached mobile viewport screenshot (image_url) alongside the sticky navigation bar and modal CSS below. Explain why the checkout drawer clips under the header at 390px width, name the stacking-context rule causing the overlap, and return a minimal patch without breaking desktop sticky behavior.

    Why it works. Combines an attached visual screenshot with code context so Claude Fable 5.2 diagnoses layout defects directly from a real capture.

    Try this prompt

Pricing

Claude Fable 5.2 API pricing

From $7.2 per million input tokens and $36 per million output tokens.

Get started

How to use the Claude Fable 5.2 API in three steps

  1. Step 1

    Pin static context at the prefix

    Place system rules, tool schemas, and reference codebases or manuscripts at the top of messages before dynamic user turns. Keeping that leading prefix byte-identical lets repeat requests settle at $0.72/1M cached reads instead of $7.20/1M.

  2. Step 2

    Match reasoning_effort to each turn

    Leave reasoning_effort at medium for balanced drafting and routine code changes, drop to low for quick extractions, and step up to high or max when diagnosing gnarly defects. Skip prompts that tell the model to think step by step so you do not inflate internal token burn.

  3. Step 3

    Reserve max_tokens for thinking

    Because internal reasoning and visible text share max_tokens (default 4,096, ceiling 64,000), raise the limit whenever you choose high or max effort. Under-sizing this cap can truncate the final answer mid-thought.

Parameter reference and code samples are in the API and Readme tabs at the top of this page.

FAQ

Claude Fable 5.2 API FAQ

About Claude Fable 5.2

Bring multi-repository engineering, visual debugging, and long-horizon literary drafting onto Anthropic Claude Fable 5.2 without rewriting your client stack. Point any OpenAI-compatible SDK to /v1/chat/completions with model ID anthropic/claude-fable-5.2/chat, keep static specs and manuscripts pinned at the top for $0.72/1M cache reads, and dial reasoning_effort from low to max per turn. Review live rates on /models/claude-fable-5-2 and compare siblings on /models/claude-fable-5 whenever a workflow calls for different output ceilings.

All Claude models11

Related models and next tasks