Anthropic launches Claude Fable 5.2 and Claude Mythos 5.2 as its highest-capability models for difficult coding, reasoning, scientific and long-horizon agent work. The November 9 release gives both models a one-million-token context window and a 128,000-token maximum output.

 

Fable and Mythos use the same underlying model but differ in safeguards and access. Fable is broadly available, while Mythos is restricted to organizations approved for cybersecurity or life-sciences work. Standard API prices fall 20 percent from the 5.1 generation to $8 per million input tokens and $40 per million output tokens.

 

The release establishes four practical terms:

  • Fable is generally available; Mythos requires verification.
  • Both models support one million input tokens.
  • Standard input and output prices fall by 20 percent.
  • An optional fast mode costs twice the standard rate.

 

Anthropic Launches Claude Fable 5.2 and Mythos 5.2

Anthropic’s announcement positions Fable 5.2 for demanding reasoning and agent sessions that may run for hours. It can also act as an advisor that smaller Claude models consult or as an orchestrator that delegates easier work while retaining responsibility for a larger task.

 

Claude Fable 5.2 accepts text and images and produces text. Adaptive thinking is always enabled, with the effort parameter controlling depth. Its reliable knowledge and training-data cutoffs are both June 2026, details that matter when a task depends on newer events or changing technical documentation.

 

The model is available through the Claude API, Amazon Web Services, Google Cloud and Microsoft Foundry. Reuters independently confirmed the launch, pricing and access split, while noting that Anthropic is broadening its model lineup ahead of a planned public offering.

 

Same Model, Different Safeguards and Access

Claude Fable 5.2 applies stronger cybersecurity and biology safeguards so Anthropic can offer it broadly. Those restrictions are expected to block many advanced cyber tasks. Claude Mythos 5.2 exposes the same underlying capabilities to approved organizations participating in Anthropic’s Cyber Verification or Life Sciences Verification programs.

 

The split lets Anthropic sell one capability tier through two access policies. A developer can use Fable for general coding, research and agent work, while a verified security team can request Mythos for vulnerability discovery and remediation. Amazon describes Mythos on Bedrock as a preview beta service rather than a generally available model.

 

Access also carries a data-handling condition. Anthropic’s documentation says Fable 5.2 requires 30-day data retention and is unavailable under zero-data-retention arrangements unless the company expressly authorizes an exception. Organizations with strict retention rules will need to resolve that term before migration.

 

More on This Story

 

Fable 5.2 Pricing and Deployment Conditions

Standard pricing drops from Fable 5.1’s $10 per million input tokens and $50 per million output tokens to $8 and $40. Cache reads cost $0.20 per million tokens, and the Batch API offers a 50 percent discount on input and output. Anthropic lists retirement no earlier than November 9, 2027.

 

The optional fast mode doubles prices to $16 per million input tokens and $80 per million output tokens. Anthropic says it can raise token-generation speed by up to 2.5 times. That is a vendor claim; the useful comparison for buyers is total task time and cost after tool calls, retries and review.

 

Fable remains more expensive than Claude Opus 5.5, which costs $4 per million input tokens and $20 per million output tokens. Anthropic recommends starting most coding and knowledge work with Opus, escalating to Fable or Mythos when a task’s complexity or duration justifies the additional cost.

 

Benchmarks Show Gains and Uneven Specialization

Anthropic reports a 49.1 percent FrontierCode 2.0 score for Fable 5.2 at Xhigh effort, compared with 32.4 percent for Fable 5.1. It also reports 69.9 percent on FrontierSWE v2. These results use Anthropic’s selected settings and should be read alongside cost, harness and effort-level differences.

 

The launch table is not uniformly favorable. Google’s Gemini 4 Carbon scores 53.9 percent on AutomationBench against Fable’s 51.4 percent, and Claude Opus 5.5 has a slightly higher head-to-head Elo score on AA-Briefcase. Anthropic also notes that competing labs used different tasks for one cybersecurity comparison.

 

Vals AI’s independent evaluation ranks Fable 5.2 first on its overall index at 71.31 percent, narrowly ahead of Claude Sonnet 5.5 at 70.85 percent and Gemini 4 Carbon at 70.45 percent. Fable led coding and finance tests at High effort, with an index cost of $17.94 per test.

 

The same evaluation exposes important weaknesses: Fable ranked twentieth of 37 on the Vals Legal Index and twenty-seventh of 84 on Tax Agent Bench. The mixed profile argues for workload-specific trials rather than treating one aggregate ranking as proof that the model is best for every professional task.

 

Illustrative feature photograph: a programmer typing code, published by Free-Photos through Pixabay and preserved on Wikimedia Commons under the CC0 1.0 public-domain dedication. The image does not depict Anthropic, Claude Fable 5.2 or Claude Mythos 5.2; it was cropped and resized.