# Claude Fable 5.2 and Mythos 5.2 Pricing and Specs

URL: https://3ammarketer.com/news/anthropic-fable-mythos-5-2-pricing-specs-benchmarks
Published: 2026-11-09

> Anthropic introduced Claude Fable 5.2 and Mythos 5.2 with $8 per million input token pricing, 1M context windows, and fast mode generation.

Anthropic launched Claude Fable 5.2 and Claude Mythos 5.2 with standard pricing set at $8 per million input tokens and $40 per million output tokens, delivering generation speeds up to 30% faster than version 5.1 models. Both releases share the same underlying model architecture, but Fable 5.2 incorporates built-in cybersecurity and biology safeguards so it can be offered to all users. [1]

## Core capabilities and model separation

While Fable 5.2 serves as a general-purpose model for demanding reasoning and long-horizon tasks, Claude Mythos 5.2 is restricted to approved organizations through Anthropic's Life Sciences and Cyber Verification Programs. Fable 5.2 features an adaptive thinking process that remains always on, operating with a default effort level set to high and a moderate comparative latency rating. [1][2]

Anthropic positions Fable 5.2 for long-horizon agentic work or for setups where evaluation runs on Claude Opus 5.5 at higher effort still fall short, as explored in our guide to [Claude Opus 5.5 vs Sonnet 5.5 pricing](/blog/claude-opus-5-5-vs-sonnet-5-5). To prevent model extraction during complex tasks, Anthropic implemented its strongest-ever protections against distillation attacks specifically for Fable 5.2, a defense topic detailed in our analysis of [AI model distillation architectures](/blog/ai-model-distillation-architectures-math-defense). [1][2]

## Performance benchmarks across coding and science

![Benchmark performance table showing Claude Fable 5.2 comparison scores across various evaluations.](https://dygolixznokmekyvajbb.supabase.co/storage/v1/object/public/images/content/31e24380e83c1688ccaf3d7c.png)

Benchmark evaluations highlight distinct strengths across coding, computer use, and domain-specific research. On agentic coding tasks within FrontierCode 2.0 at Xhigh effort, Fable 5.2 scores 49.1%. For OSWorld 2.1 partial offline computer use, Fable 5.2 reaches 89.3%. [1]

Agentic coding performance on FrontierCode 2.0 at Xhigh effort (%), comparing Claude Fable 5.2 to other models. [1]

- Fable 5.2: 49.1 %
- Fable 5.1: 32.4 %
- Opus 5.5: 45.5 %
- GPT-6 Astra: 42.8 %
- Gemini 4 Carbon: 41.1 %

In scientific agent evaluations on Terminal-Bench-Science 0.1, Fable 5.2 achieves 74.6%, while Mythos 5.2 reaches 75.3%. Mythos 5.2 also records 76.3% on the CyScenarioBench cybersecurity evaluation and 85.1% on the 24-hour De novo binder design protein design benchmark. In multi-agent pipelines, Fable 5.2 can operate as an Advisor that smaller models consult for difficult problems, or as an Orchestrator delegating simpler tasks to smaller models, matching patterns found in [AI agent orchestration](/blog/ai-agent-orchestration). [1]

Scientific agent evaluation scores on Terminal-Bench-Science 0.1 (%) for Claude Fable 5.2, Mythos 5.2, and competitor models. [1]

- Fable 5.2: 74.6 %
- Mythos 5.2: 75.3 %
- Fable 5.1: 52.6 %
- Opus 5.5: 58.7 %
- GPT-6 Astra: 68.1 %
- Gemini 4 Carbon: 64.8 %

## Pricing, fast mode, and prompt caching

Anthropic offers both standard generation and an optional fast mode that accelerates token throughput by up to 2.5 times. Standard pricing is $8 per million input tokens and $40 per million output tokens, while fast mode costs $16 per million input tokens and $80 per million output tokens. Prompt caching reduces intake fees to $0.20 per million tokens for reading and $10 per million tokens for writing. [1][3]

Standard versus Fast mode pricing for Claude Fable 5.2 ($ per 1M tokens). [1]

- Std Input: 8 $
- Std Output: 40 $
- Fast Input: 16 $
- Fast Output: 80 $

## Technical specifications for integration

Engineers integrating Fable 5.2 call the model using the Claude API ID claude-fable-5-2. The model accepts a context window of 1M tokens and supports maximum output generations up to 128K tokens, as examined in our overview of [what is a context window in AI](/blog/what-is-a-context-window-in-ai). [2]

The reliable knowledge cutoff for Fable 5.2 is June 2026. Anthropic has committed to maintaining support for Claude Fable 5.2 with a retirement date scheduled no sooner than November 9, 2027. [2]

Our read: For long-horizon agentic work, deploy Fable 5.2 in standard mode at $8 per million input tokens. Enable the $16 fast mode only on interactive user-facing workflows where a 2.5 times generation speedup justifies doubling token costs.

Limit: These sources do not publish token-per-second metrics for standard or fast mode, nor do they detail the specific qualification criteria required for access to Mythos 5.2 under the Cyber Verification Program.

## Sources

1. [Introducing Claude Fable 5.2 and Claude Mythos 5.2](https://www.anthropic.com/claude-fable-and-mythos-5-2)
2. [Models overview - Claude Platform Docs](https://platform.claude.com/docs/en/models/overview)
3. [Claude Platform | Claude by Anthropic](https://claude.com/platform/api)
