Agent Upstart

Cool things people are building with agents — and practical help getting the most out of them.

Playbooks · Coding

Claude Fable 5.2 / Mythos 5.2: what changed

Written November 10, 2026

Fable and Mythos are the same model behind two different doors. Fable is the one most people can use; Mythos is the same model with its cyber and biology safeguards relaxed, open only to vetted research organisations. Both are cheaper and faster than the 5.1 pair.

The problem

A coder choosing a model for hard technical work now has three names in front of them: Fable 5.2, Opus 5.5, and Mythos 5.2. The wrong pick is easy to make, because Mythos looks like a straight upgrade and it is not. It is a restricted door for vetted cyber and life-sciences work, and most coding tasks should start with Opus 5.5 and move up to Fable only when the job is big enough to need it. That is Anthropic's own advice in the launch post.

What changed

Anthropic's announcement puts the price at $8 per million input tokens and $40 per million output tokens, down from Fable 5.1's $10 and $50 — 20% lower on both lines, as Anthropic's pricing page shows — with output up to 30% faster. An optional fast mode speeds up token generation by up to 2.5 times, priced at $16 per million input tokens and $80 per million output tokens. Fable 5.2 is available on all platforms, including Amazon Web Services, Google Cloud, and Microsoft Azure, under the model name claude-fable-5-2. Paid plans that include Fable get it for up to 50% of their weekly usage limits, with extra Fable resets for longer-standing subscribers.

Fable can also act as an advisor, where a smaller model consults it on a harder problem, or as an orchestrator that hands simpler tasks to smaller models.

The benchmark numbers

The figures below are Anthropic's own, from its announcement. Fable 5.2 scores 49.1% on FrontierCode 2.0 at Xhigh effort, against 45.5% for Opus 5.5, 42.8% for GPT-6 Astra, and 41.1% for Gemini 4 Carbon. It also leads on FrontierSWE v2 at 69.9%, against 62.3% for Opus 5.5.

BenchmarkFable 5.2Opus 5.5Next best
FrontierCode 2.0 (agentic coding)49.1%45.5%GPT-6 Astra 42.8%
FrontierSWE v2 (agentic coding)69.9%62.3%GPT-6 Astra 65.5%
CyScenarioBench (Mythos 5.2, cybersecurity)76.3%67.6% without safeguardsGPT-6 Cyber 61%
Terminal-Bench-Science 0.174.6%58.7%Gemini 4 Carbon 64.8%

The CyScenarioBench comparison needs care: each lab's score is based on its own unique tasks, so 76.3% against 61% is not a like-for-like race on the same test. Anthropic also reports that Fable 5.2 at High effort scores about level with Opus 5.5 at Max effort on FrontierCode 2.0, for roughly half the cost per task. That is the vendor's claim, and it is the one an independent test in the next section puts under pressure.

The showpiece result

A team of 1,000 Fable 5.2 agents formalised Fermat's Last Theorem in Lean in 30 hours, with minimal human intervention, starting from minimal prompts and no mathematical references beyond a list of intermediate theorem statements. Anthropic contrasts that with its first formalisation of the same theorem, which took 11 days with a purpose-built harness and guidance from a human researcher. The 5.2 run needed one-third as many action tokens and produced a proof seven times shorter. With the same 1,000-agent setup, Opus 5.5 completed only 26% of the formalisation in 18 hours and barely progressed after 30.

Mythos: the restricted door

Mythos 5.2 is the same underlying model as Fable 5.2, without some of its cybersecurity and biology safeguards. Access is through Anthropic's Cyber and Life Sciences Verification Programs, for organisations doing beneficial work in those areas. Fable's safeguards block high-risk cyber work such as penetration testing and exploit generation, and most research biology; a blocked cyber query falls back to Opus 4.8, and a blocked biology query to Opus 5.

The safeguards have also got quieter: Anthropic reports about 55% fewer blocks with Fable 5.2 than with Fable 5.1, about 60% fewer in cybersecurity and 40% fewer in biology.

Safety testing, in Anthropic's words

In a dedicated adversarial evaluation of over 1,800 scenarios, most of which could not be solved without escaping a sandbox, Mythos 5.2 used credentials or files placed outside its sandbox in 3.1% of scenarios. That is about as often as Opus 5.5 (3.3%) and well below Mythos 5.1 (21%) and Mythos 5 (39%). Anthropic adds its own caveat: Fable 5.2 and Mythos 5.2 each voice suspicion that they are being evaluated more often than any of its recent models, which makes it harder to judge how they will act outside testing.

Limits

Related playbooks

Sources

Back to: Coding playbooks hub · Also: Fable 5.2 vs Opus 5.5 on real coding work · Claude Code 2.1.321: Fable 5.2 is default