Anthropic announces Claude Fable 5.2 and Mythos 5.2: higher agent benchmark scores
On November 9, 2026 Anthropic announced Claude Fable 5.2 and Claude Mythos 5.2. The company says they are the same underlying model: Fable ships with cyber and biology safeguards and is open to all users, while Mythos is reserved for verified organizations. Output is up to 30% faster than Fable 5.1, and Fable 5.2 is available on AWS, Google Cloud and Azure among other platforms. The benchmark figures are Anthropic's own.
One model under two names, with access split by safeguards
In its November 9, 2026 announcement Anthropic introduced Claude Fable 5.2 and Claude Mythos 5.2 and says they are the same underlying model. The difference is the safety layer: Fable carries safeguards for cyber and biology topics and is available to everyone, while Mythos goes to verified organizations, which apply through the Cyber Verification Program or the Life Sciences Verification Program, both of which Anthropic says it has expanded.
Fable 5.2 is available on all platforms, including Amazon Web Services, Google Cloud and Microsoft Azure, and the API model ID is claude-fable-5-2. The announcement names no geographic restriction. It continues the story we covered in our Fable 5.1 article.
Speed, benchmark results and cost per effort level
Anthropic says output is up to 30% faster than Fable 5.1 and Mythos 5.1. Fast mode raises token generation speed by up to 2.5x, but it is offered only on Claude Code, the Claude Platform, Claude models on Google Cloud and Microsoft Foundry. For Mythos, fast mode is limited to organizations in the verification programs.
The benchmark table compares Fable 5.2, Fable 5.1, Opus 5.5, GPT-6 Astra and Gemini 4 Carbon. On AutomationBench Fable 5.2 scores 51.4 (Fable 5.1: 31.4), but Gemini 4 Carbon scores higher at 53.9, so Fable 5.2 does not lead that one. On the Remote Labor Index Fable 5.2 reaches 41.3% at High effort versus 17.9 for Fable 5.1. On OSWorld 2.1 (offline, partial) it is 89.3 against 83.5, and on FrontierSWE v2 69.9 against 56.3. On FrontierCode 2.0 at Xhigh effort it scores 49.1, against 32.4 for Fable 5.1 and 45.5 for Opus 5.5.
Anthropic also reports that Fable 5.2 at High effort scores 45.3% on FrontierCode 2.0, roughly level with Opus 5.5's best score (45.5%) at about half the cost per task. According to the page, scores are with safeguards on and at each model's highest available effort unless stated otherwise.
A jump on agent-style work, but the measurements come from the maker
Most of these results measure agent-style work such as automation, computer use and coding. The gaps against Fable 5.1 look large, yet every figure is Anthropic's own report, not independent verification. Our sources are also limited to two official Anthropic pages.
The second notable part is the division of roles. Anthropic positions Opus 5.5 as the main daily tool for most coding and knowledge work, with users moving up to Fable or Mythos, or down to Sonnet 5.5 or Haiku 5.5, as needed. Fable 5.2 can act as an advisor that Haiku, Sonnet or Opus consult, or as an orchestrator that delegates to smaller models.
On safeguards the page says Fable 5.2's classifiers block high-risk cyber activities such as penetration testing and exploit generation, as well as most tasks in research biology. Blocked cyber queries fall back to Opus 4.8 and blocked biology queries to Opus 5, and overall blocks are about 55% fewer than with Fable 5.1. As a launch promotion, Fable is included for up to 50% of weekly usage limits on Pro and standard Team seats.
The announcement does not mention Türkiye, so this section is commentary
The announcement page says nothing about Türkiye or Turkish businesses and names no geographic restriction. What follows is not drawn from the source; it is UNALSOFT commentary.
Our view: if you are considering agents for customer support, bookings or internal operations, the practical lesson is not to tie yourself to a single model. Anthropic itself tiers its models by job, and having the same model offered through major cloud providers may ease integration with infrastructure you already use. Terms of use, access and cost should be verified separately on your own account; this article makes no promise about them.
The UNALSOFT view
We turn this story into one question: which step in a workflow genuinely needs which model tier? Starting with smaller models and moving only the hard steps upward is a measurable approach, compared with handing every job to the strongest model. That requires first pinning down what the job is, who catches mistakes and where human sign-off belongs. This is the frame we use in our agentic AI work. Because the benchmark figures come from the maker, we suggest a small trial on your own data before deciding.
Sources
Anthropic, Introducing Claude Fable 5.2 and Claude Mythos 5.2 · Anthropic News, announcement index
Is your customer flow ready for AI agents?
A short conversation is enough to review your model choices and agent workflows together.