Claude Fable 5.2 Guide: Pricing, Prefix Checks, and Migration

Learn what changed with Anthropic's Claude Fable 5.2 release, covering adaptive thinking rules, fast mode pricing, prefix check enforcement, and tool migrations.

By ailog · Published Nov 15, 2026 · 6 min read · How we work
In short
  • Claude Fable 5.2 and Claude Mythos 5.2 deliver 1M context tokens and 128k maximum output tokens with mandatory adaptive thinking.
  • Input and output rates decrease to $8 and $40 per million tokens, while an optional fast mode offers up to 2.5x throughput at $16 and $80.
  • Strict prefix checking applies universally to all accounts, returning HTTP 400 errors whenever prompts or tools are edited before prior thinking turns.
  • Computer use transitions to computer_toolset_20260801 across direct endpoints, while agent advisor workflows restrict executors to 5.2 series models.
  • Data retention remains fixed at 30-day logging, with zero data retention unavailable absent explicit Anthropic authorization.

Claude Fable 5.2 Architecture and Core Capabilities

On 2026-11-09, Anthropic announced Claude Fable 5.2 under the model identifier claude-fable-5-2, alongside its restricted twin Claude Mythos 5.2 (claude-mythos-5-2). Both variants target prolonged reasoning chains and long-horizon autonomous agents. While Claude Fable 5.2 is open for public developer integration, Claude Mythos 5.2 requires validation through specialized initiatives such as the Cyber Verification Program.

Each model ships with a standard 1M token context window and generates up to 128k output tokens per turn. Unlike earlier generations where thinking was configurable via discrete token allotments, both 5.2 releases enforce always-on adaptive thinking. Claude determines internal reasoning depth on a per-request basis. Attempting to disable reasoning by passing thinking: {"type": "disabled"} or supplying manual budgets through "budget_tokens" causes the API to reject the call with an HTTP 400 error.

Engineering teams adopting claude-fable-5-2 gain broader multi-cloud access on launch day. The model is deployed across the primary Claude API, Amazon Bedrock (bedrock-runtime), Claude Platform on AWS, managed Google Cloud endpoints (both Anthropic managed and Google managed), and Claude in Microsoft Foundry. In addition, Managed Agents Dreams now incorporates Claude Fable 5.2 into its research preview environment.

Pricing Adjustments and Fast Mode Execution

The release introduces lower baseline rates alongside an accelerated execution tier. Claude Fable 5.2 charges $8 per million input tokens and $40 per million output tokens. This reduces token expenses compared to Claude Fable 5.1, which billed at $10 for inputs and $50 for outputs per million tokens. Teams balancing high-volume agent pipelines can evaluate overall workload budgets on the /tools/llm-api-cost-calculator/.

Model Identifier Input Rate (per 1M) Output Rate (per 1M) Context Limit Max Generation
claude-fable-5-2 $8 $40 1M tokens 128k tokens
claude-fable-5-2 (fast mode) $16 $80 1M tokens 128k tokens
claude-fable-5.1 $10 $50 1M tokens 128k tokens
claude-opus-5.5 $4 $20 1M tokens 128k tokens
claude-sonnet-5.5 $2 $10 1M tokens 128k tokens

For interactive agent loops requiring minimum response delay, Anthropic launched fast mode for Claude Fable 5.2 on the Claude API Messages endpoint. Fast mode yields up to 2.5x higher output tokens per second relative to standard generation. Selecting this speed doubles API rates, invoicing input tokens at $16 per million and output tokens at $80 per million. Developers can review catalog comparisons on /tools/llm-api-pricing/.

curl https://api.anthropic.com/v1/messages \
  -H "x-api-key: $ANTHROPIC_API_KEY" \
  -H "anthropic-version: 2026-10-21" \
  -H "content-type: application/json" \
  -d '{
    "model": "claude-fable-5-2",
    "max_tokens": 16000,
    "speed": "fast",
    "messages": [
      {"role": "user", "content": "Analyze system logs for security anomalies."}
    ]
  }'

Preserved Thinking and Universal Prefix Check Enforcement

Managing reasoning blocks across multi-turn sessions requires strict structural adherence under Fable 5.2. Both Claude Fable 5.2 and Claude Mythos 5.2 strictly enforce the prefix check across every developer account, irrespective of creation date. Under Claude Fable 5.1, prefix validation errors were default only for accounts registered on or after August 31, 2026. With Fable 5.2, all legacy exemptions are removed.

If client code modifies the system prompt, edits definitions in tools, or alters earlier message blocks before passing back a preserved thinking block from a subsequent conversational turn, the API rejects the payload with an HTTP 400 error. Cryptographic signatures attached to each thinking block ensure that prior context remains intact. Teams migrating stateful architectures must confirm that conversation history remains append-only when preserving reasoning chains.

  • Inbound Compatibility: Claude Fable 5.2 and Claude Mythos 5.2 can read reasoning blocks emitted by earlier models, including Claude Opus 5.5, Claude Sonnet 5.5, and Claude Haiku 5.5, when running on the Claude API or Google-managed Cloud endpoints.
  • Outbound Isolation: No external model can consume thinking blocks produced by Claude Fable 5.2 or Mythos 5.2. When moving a dialogue to Claude Opus 5.5 or Claude Fable 5.1, the API automatically strips those blocks while allowing the request to proceed.
  • Tenant Boundary: Thinking blocks created by Fable 5.2 or Mythos 5.2 can only be reused within the origin account or an explicitly linked organization account.

Tool Ecosystem Changes: Computer Use and the Advisor Tool

Integration with external tools changes significantly when upgrading workloads to claude-fable-5-2. For automated operating system tasks, the model mandates the newer computer_toolset_20260801 specification. Submitting the legacy computer_20251124 identifier returns an HTTP 400 validation error on the Claude API and partner platforms, with Amazon Bedrock remaining the sole environment where the legacy tool continues functioning.

For agentic patterns utilizing the beta advisor tool, model compatibility rules have tightened. When an agent designated as the executor runs on Claude Fable 5.2 or Claude Mythos 5.2, its advisor model parameter must also be set strictly to Claude Fable 5.2 or Claude Mythos 5.2. Configuring any third model as the advisor triggers a 400 failure. Conversely, executors running on other Claude models can assign Claude Fable 5.2 as their advisor without issue.

Executor Model Allowed Advisor Models Behavior on Other Advisors
claude-fable-5-2 claude-fable-5-2, claude-mythos-5-2 HTTP 400 validation error
claude-mythos-5-2 claude-fable-5-2, claude-mythos-5-2 HTTP 400 validation error
claude-opus-5.5 claude-fable-5-2 and other models Accepted
claude-sonnet-5.5 claude-fable-5-2 and other models Accepted

Cloud Infrastructure, Routing, and Retention Boundaries

Cloud deployments on Amazon Bedrock interact with model IDs via cross-region inference profiles. Rather than providing regional on-demand throughput under bare model IDs, Bedrock requires routing through inference profiles. When utilizing the bedrock-runtime endpoint, developers must prepend global or regional profile strings to the base identifier anthropic.claude-fable-5-2.

Bedrock Model Name Base Identifier Supported Inference Profiles
Claude Fable 5.2 anthropic.claude-fable-5-2 global, us
Claude Mythos 5.2 anthropic.claude-mythos-5-2 global, us
Claude Opus 4.6 anthropic.claude-opus-4-6-v1 global, us, eu, jp, apac

Data privacy requirements remain identical to Fable 5.1. Both Claude Fable 5.2 and Claude Mythos 5.2 enforce 30-day data retention by default across all standard API tiers. Zero data retention policies cannot be enabled for these models unless an organization has obtained explicit, documented authorization directly from Anthropic.

Migration Verification Checklist

Before modifying production configurations to point to claude-fable-5-2, engineering teams should review their request pipelines against these operational criteria to prevent run-time errors.

  • Audit conversation history buffers: Confirm client code does not mutate system blocks, inject dynamic tool schemas, or rewrite past turns when passing back signed reasoning blocks.
  • Verify thinking configuration: Remove any thinking: {"type": "disabled"} or budget_tokens values from API requests to avoid immediate HTTP 400 rejections.
  • Migrate computer use schemas: Replace legacy declarations of computer_20251124 with computer_toolset_20260801 across non-Bedrock deployments.
  • Check advisor tool pairings: Ensure that any executor agent assigned to claude-fable-5-2 pairs exclusively with an advisor running Fable 5.2 or Mythos 5.2.
  • Review latency and cost demands: Enable speed: "fast" without beta headers for throughput-critical workflows while planning for the adjusted $16 / $80 token pricing.
  • Confirm data retention posture: Verify that internal data governance permits the mandatory 30-day retention window enforced on both 5.2 models.
Sources
  1. Anthropic Claude Platform release notes — 2026-11-09
  2. Context windows
  3. Thinking
  4. Claude on Amazon Bedrock (bedrock-runtime)

Related