Claude Fable 5.2 Explained
Capabilities, access boundaries, pricing, and breaking API changes for Anthropic's flagship intelligence model.
Claude Fable 5.2 is Anthropic's flagship artificial intelligence (AI) model, released on November 9, 2026 under the model identifier claude-fable-5-2. The model features a 1 million token context window with always-on adaptive thinking, priced at $8 per million input tokens and $40 per million output tokens. Access is open to all Claude Application Programming Interface (API) customers and to users on paid consumer tiers. Claude Mythos 5.2 serves as the restricted-access twin for verified organizations.
Anthropic provides Claude Fable 5.2 across major cloud platforms including Amazon Web Services (AWS) Bedrock, Google Cloud, and Microsoft Azure Foundry. The model costs 20 percent less than Claude Fable 5.1 and generates responses up to 30 percent faster according to Anthropic's launch announcement. It also operates under Anthropic's Covered Model policy, which establishes baseline safety reviews and data retention requirements.
Deploying Claude Fable 5.2 requires examining its technical specifications, breaking API changes, and positioning within Anthropic's broader lineup. This reference outlines the model's access boundaries, benchmark results, pricing structures, and retirement commitments to assist technical teams evaluating production workloads.
What Claude Fable 5.2 Is and Who Can Use It
Claude Fable 5.2 is Anthropic's primary flagship model for complex coding, scientific reasoning, and knowledge work, available to all Claude API accounts and paid subscription tiers.
Anthropic launched the model on November 9, 2026 using the API identifier claude-fable-5-2. Claude Fable 5.2 and Claude Mythos 5.2 share the same underlying architecture and capabilities, but Anthropic applies different distribution policies to each model. Claude Fable 5.2 incorporates cybersecurity and biological safeguards that allow general commercial availability. In contrast, Anthropic restricts Claude Mythos 5.2 strictly to organizations verified through specialized initiatives, such as the Cyber Verification Program and the Life Sciences Verification Program. Teams evaluating the separation between the two distributions can consult the detailed Claude Fable 5.2 vs Mythos 5.2 guide.
Deployment for Claude Fable 5.2 spans the Claude API, Amazon Bedrock (under the model identifier anthropic.claude-fable-5-2), Claude Platform on AWS, Claude Models on Google Cloud (both Anthropic-managed and Google-managed), and Claude in Microsoft Azure Foundry. Across consumer and workplace plans, access depends on tier entitlements published on the Anthropic pricing page. The Free plan does not include Claude Fable 5.2. Pro subscriptions ($20 monthly, or $17 monthly billed annually) access Claude Fable 5.2 using usage credits. Max costs $100 monthly and Ultra costs $200 monthly. Both include Claude Fable 5.2 for up to 50 percent of weekly limits. Team plans include Claude Fable 5.2 up to 50 percent of weekly limits on premium seats, while Enterprise plans provide full access.
For the initial launch period, Anthropic introduced a dated promotional policy documented in support article 17506494. From launch on November 9, 2026 through November 16, 2026 at 11:59 pm Pacific Time (PT), Pro subscribers and standard seats on Team plans get all Fable models up to 50 percent of weekly limits. They also get one usage limit reset that expires November 16, 2026 at 11:59 pm PT. After that promotional window closes, standard Team seats and Pro accounts return to usage credits.
Claude Fable 5.2 Facts at a Glance
Claude Fable 5.2 has a 1 million token context window. It allows 128,000 maximum output tokens, and adaptive thinking is always on.
| Specification | Value | Primary Source |
|---|---|---|
| Context window | 1M tokens (default and maximum) | Anthropic Models Overview |
| Maximum output tokens | 128,000 tokens | Anthropic Models Overview |
| Default effort setting | High (API default); Medium (Claude apps and Claude Code) | What's New in Claude Fable 5.2 |
| Token pricing | $8 input / $40 output per million tokens | Anthropic Platform Pricing |
| Fast mode pricing | $16 input / $80 output per million tokens | Anthropic Platform Pricing |
| Training knowledge cutoff | June 2026 | Claude Fable 5.2 Overview |
| Host platforms | Claude API, AWS Bedrock, Google Cloud, Microsoft Foundry | Anthropic Announcement |
| Retirement floor | Not sooner than November 9, 2027 | Anthropic Model Deprecations |
| Covered Model status | Designated Covered Model (30-day retention default) | Covered Models Documentation |
Unlike models that charge higher rates for long contexts, Claude Fable 5.2 bills the full 1 million token window at standard token rates. See the complete Claude Fable 5.2 pricing guide for cache, read and Batch API details.
What Changed From Claude Fable 5.1
Upgrading from Claude Fable 5.1 to Claude Fable 5.2 delivers 20 percent lower token pricing and up to 30 percent faster output alongside four breaking API changes.
Anthropic reports that Claude Fable 5.2 generates text up to 30 percent faster than its predecessor. Based on internal cost modeling from Anthropic's published rates, standard pricing drops by 20.0 percent on base input ($8 versus $10 per million tokens) and base output ($40 versus $50 per million tokens). That identical 20.0 percent discount applies across 5-minute prompt cache writes ($10 versus $12.50), 1-hour cache writes ($16 versus $20), cache reads ($0.20 versus $0.25), and asynchronous Batch API jobs ($4 input and $20 output versus $5 and $25). Teams on the prior version can read Claude Fable 5.1 explained.
Engineering teams migrating existing codebases must account for several structural differences documented in Anthropic's migration guide:
- Enforced prefix validation: Thinking block verification applies across all accounts. Altering the system prompt, available tools, or previous conversation history while passing back a Claude Fable 5.2 thinking block returns an HTTP 400 error unless developers enable the beta header and set prefix_mismatch_behavior to drop_block.
- Computer use toolset migration: The legacy computer_20251124 tool identifier returns an HTTP 400 error on every platform except Amazon Bedrock. Callers must update integrations to computer_toolset_20260801.
- Advisor tool pairing restrictions: In the beta advisor tool, a Claude Fable 5.2 executor agent accepts only Claude Fable 5.2 or Claude Mythos 5.2 as an advisory model.
- Thinking-block readability: Claude Fable 5.1 cannot read Claude Fable 5.2 thinking blocks. Always-on adaptive thinking and the forced tool choice restriction carry over unchanged from Claude Fable 5.1.
What Anthropic Reports It Does Well
Vendor-reported evaluations published by Anthropic show Claude Fable 5.2 achieving leading marks across major coding, scientific reasoning, and autonomous agent benchmarks.
In the benchmark suite published in Anthropic's announcement, Anthropic evaluated Claude Fable 5.2 against Claude Fable 5.1, Claude Opus 5.5, OpenAI's GPT-6 Astra, and Google's Gemini 4 Carbon. All Anthropic models were evaluated with safeguards active at each model's highest available effort unless specifically noted. Full score breakdowns are analyzed on the Claude Fable 5.2 benchmarks tracker. Across 12 non-Mythos benchmark rows evaluated by Anthropic, Claude Fable 5.2 achieved the highest score among listed models on 10 rows and scored higher than Claude Fable 5.1 on all 12 rows.
On coding and software engineering, Anthropic reports Claude Fable 5.2 scored 49.1 percent on FrontierCode 2.0 at extra-high effort (compared to 32.4 percent for Fable 5.1, 45.5 percent for Opus 5.5, and 42.8 percent for GPT-6 Astra) and 69.9 percent on FrontierSWE v2 (compared to 56.3 percent for Fable 5.1 and 65.5 percent for GPT-6 Astra). On agentic science, it scored 74.6 percent on Terminal-Bench-Science 0.1 (compared to 52.6 percent for Fable 5.1 and 68.1 percent for GPT-6 Astra), while Claude Mythos 5.2 reached 75.3 percent. On operating system control, it achieved 89.3 percent on OSWorld 2.1 offline. It recorded 74.2 percent on Humanity's Last Exam (HLE) with tools and 82.1 percent on WANDR relaxed. Claude Fable 5.2 trailed other models on two evaluations: Gemini 4 Carbon led AutomationBench (53.9 percent versus 51.4 percent for Fable 5.2). Claude Opus 5.5 led the AA-Briefcase v1.1 Elo rating (1800 Elo versus 1785 Elo for Fable 5.2).
Anthropic also published task-level efficiency data. At high effort, Claude Fable 5.2 scored level with Claude Opus 5.5 at maximum effort on FrontierCode 2.0 (45.3 percent versus 45.5 percent) for roughly half the cost per task. At medium effort on WANDR, it matched Claude Opus 5.5's top score for about 40 percent of the cost per task. On HLE at low effort, it matched Claude Opus 5.5's top score for under one-tenth the cost. Anthropic's announcement reports that a team of agents on an earlier Claude Fable 5.2 version migrated a production service from Python to Rust, starting about 26 times faster. It also reports that one engineer's prototype operating system used over 1,000 subagents over eight days. Prerelease testing also formalized the proof of Fermat's Last Theorem in 30 hours. These anecdotes come from Anthropic's prerelease use. Anthropic also reported that over 90 percent of changes merged into Anthropic's core codebase are written by Claude with human guidance.
Regarding model alignment, Anthropic's announcement states that high-stakes risks from misalignment remain low, that cybersecurity and biology safeguards are as strong as those in Fable 5.1 while generating fewer unnecessary blocks. Anthropic also reports enhanced protections against distillation attacks. Anthropic notes details are in the Claude Fable 5.2 and Mythos 5.2 System Card.
For the current OpenAI rival, compare Claude Fable 5.2 with GPT-6.2 Astra while keeping the reported benchmark conditions separate.
Where Claude Fable 5.2 Fits Against Claude Opus 5.5
Anthropic recommends that development teams start new coding and knowledge tasks on Claude Opus 5.5 before upgrading workloads to Claude Fable 5.2.
Anthropic positions Claude Opus 5.5 as its primary daily model for general development work. Billed at $4 per million input tokens and $20 per million output tokens, Claude Opus 5.5 costs half as much as Claude Fable 5.2. Readers can review Claude Opus 5.5 explained for a technical overview of that architecture. Anthropic's official models overview specifies that teams should only move production tasks to Claude Fable 5.2 when evaluation suites executed on Claude Opus 5.5 at higher effort levels fail to satisfy project requirements.
Anthropic's documentation describes Claude Fable 5.2 as able to act as an Advisor or an Orchestrator. Under an Advisor pattern, smaller endpoints such as Claude Sonnet 5.5 ($2 input / $10 output) or Claude Haiku 5.5 ($0.10 input / $0.50 output) consult Claude Fable 5.2. Under an Orchestrator pattern, Claude Fable 5.2 delegates execution to smaller instances. A full operational comparison is detailed in the guide on whether Claude Fable 5.2 is worth it.
How Long Anthropic Keeps It Available
Anthropic lists a tentative retirement date of not sooner than November 9, 2027 for Claude Fable 5.2 under its published API deprecation schedule.
In the official model deprecations register, Anthropic lists Claude Fable 5.2 with Active status and a tentative retirement date of 'Not sooner than November 9, 2027'. This date represents an availability floor rather than an announced shutdown. Under Anthropic's policy, publicly released models remain supported until at least that date. Anthropic commits to providing a minimum of 60 days of advance notice before turning off any production model ID.
This lifecycle timeline mirrors Anthropic's policy across current frontier systems. Claude Fable 5.1 and Claude Mythos 5.1 hold retirement floors not sooner than September 1, 2027. Claude Opus 5.5 holds a floor not sooner than September 22, 2027. Teams maintaining compliance records can monitor lifecycle windows on the Anthropic model deprecations and retirement dates tracker.
Who Should Not Start With Claude Fable 5.2
Organizations requiring unconditional zero data retention or high-throughput conversational routing should not deploy Claude Fable 5.2 as their starting model.
Anthropic classifies Claude Fable 5.2 as a Covered Model under its Covered Models documentation. By default, Covered Models require a 30-day minimum retention period for automated safety reviews across all endpoints. Standard Zero Data Retention (ZDR) agreements do not apply to Covered Models in standard API workspaces, Claude Enterprise setups, or third-party cloud deployments. While Anthropic provides a temporary transition path for eligible enterprise organizations to request ZDR for internal business applications pending the staged rollout of Enterprise Frontier Safeguards (EFS), organizations subject to regulatory policies that forbid any vendor data retention cannot use Claude Fable 5.2 out of the box. Teams managing protected health information (PHI) should review the analysis on whether Claude Fable 5.2 is HIPAA compliant to verify Business Associate Agreement (BAA) prerequisites.
Budget-constrained teams and consumer users on the Free tier also cannot run Claude Fable 5.2. Simple high-volume jobs do not warrant an $8 input and $40 output rate card. For these tasks, Claude Sonnet 5.5 or Claude Haiku 5.5 deliver higher throughput at substantially lower cost.
For engineering teams already running Claude Fable 5.1, Claude Fable 5.2 represents a faster and cheaper replacement once developers test the four breaking API changes. For new agentic implementations, evaluate baseline error rates on Claude Opus 5.5 before upgrading production workloads to Claude Fable 5.2.
Frequently Asked Questions
- Claude Fable 5.2 was released on November 9, 2026 across the Claude API, consumer subscription plans, Amazon Bedrock, Google Cloud, and Microsoft Azure Foundry.
- Claude Fable 5.2 costs $8 per million input tokens and $40 per million output tokens for standard API calls, with optional fast mode billed at $16 input and $80 output per million tokens as detailed on the Claude Fable 5.2 pricing guide.
- Anthropic published details in the Claude Fable 5.2 and Mythos 5.2 System Card.
- Claude Fable 5.2 is Anthropic's flagship model for high-difficulty coding and agentic workflows, whereas Claude Opus 5.5 costs half as much ($4 input / $20 output) and serves as Anthropic's recommended default model for daily development.
- Claude Fable 5.2 and Claude Mythos 5.2 share the same underlying architecture and capabilities, but Mythos 5.2 is restricted to organizations verified through Anthropic's specialized cybersecurity and life sciences partnership programs.
- Claude Fable 5.2 is available on the Claude API, Claude Pro, Max, Ultra, Team, and Enterprise plans, Amazon Bedrock, Google Cloud, and Microsoft Azure Foundry.
Plan an AI Model Architecture Review
Evaluate how Claude Fable 5.2 fits your technical requirements, latency targets, and data compliance mandates. Layer3Labs reviews endpoint architectures to identify cost savings and upgrade paths across enterprise AI deployments.
Book a Consultation