Home  /  Blog  /  Fable 5.2 and Mythos 5.2
NEWS / MODELS

Fable 5.2
and Mythos 5.2.

Yesterday, 9 November 2026, Anthropic released Claude Fable 5.2 and Claude Mythos 5.2. They are the same model with two different ways in: Fable for everyone, with safeguards in cybersecurity and biology, and Mythos only for vetted organizations. Here is what's new in Claude Fable 5.2, what it costs, where you can use it and what we still don't know.

BY SKOA~5 MIN READ
Claude Fable 5.2 and Claude Mythos 5.2: one model, two ways in. Fable for everyone, Mythos for vetted organizations.
01 / ONE MODEL, TWO DOORS.
9 Novrelease date
$8 / $40per million tokens
1Mtoken context (Fable)
up to 30%faster than 5.1
01 /

What's out.

On 9 November 2026 Anthropic announced two models at once: Claude Fable 5.2 and Claude Mythos 5.2. They replace Fable 5.1 and Mythos 5.1. According to Anthropic, both communicate more clearly than the previous version, improve in coding, reasoning and vision, and generate text up to 30% faster.

Fable 5.2 is available today on every platform. Mythos 5.2 is only open to organizations accepted into its verification programs.

02 /

One model, two ways in.

The first thing to understand is that these are not two different models. The announcement says it plainly: Fable and Mythos are the same underlying model. The difference is who can use it and with what limits.

Fable 5.2Mythos 5.2
WhoAnyone: Claude apps, API and cloudsVetted organizations
SafeguardsStrong in cybersecurity and biologyFewer restrictions in those two fields
Built forThe hardest, longest technical workCyber defense and life sciences research
AccessDirectlyCyber Verification Program or Life Sciences Verification Program

If you ask Fable 5.2 a cybersecurity or biology question that falls in the blocked zone, Anthropic answers with an older model instead of simply refusing. It also says that, in early testing, Fable 5.2 blocks 55% fewer harmless requests than Fable 5.1.

Claude Enterprise customers already use Mythos 5.2 indirectly through Claude Security, which now runs on this model.

03 /

What's new in Claude Fable 5.2.

Every figure in this section comes from Anthropic. These are internal or third-party tests chosen by the company itself, so read them for what they are: the maker's version.

  1. Agentic coding: 49.1% on FrontierCode 2.0 at Xhigh effort, against 45.5% for Opus 5.5 and 32.4% for Fable 5.1. Anthropic says it is the best score of any released model it has tested.
  2. Computer use: 89.3% on OSWorld 2.1.
  3. Robotics and hardware: 64.9% on RoboDojo Sim, against 27.6% for Fable 5.1, and 83.5 on BenchCAD 2.1 at Max effort, against 65.3 for Opus 5.5.
  4. Cybersecurity (Mythos 5.2): on the internal CVE Rediscovery test it found 10.2% more vulnerabilities than Opus 5.5 and 14.4% more than Mythos 5.1.
  5. Math: Anthropic reports that about 1,000 Fable 5.2 agents formalized Fermat's Last Theorem in Lean in 30 hours. It is the most striking claim in the announcement and we have not been able to check it.
  6. Fast mode: optional, generates tokens up to 2.5 times faster and costs twice as much.
Where it doesn't improve much

Anthropic admits that on AI research tasks, such as optimizing kernels or designing model architectures, both models perform roughly at the level of Opus 5.5.

04 /

Pricing and availability.

ItemFable 5.2Mythos 5.2
Input / output (per million tokens)$8 / $40$8 / $40
Batch$4 / $20$4 / $20
Fast mode$16 / $80$16 / $80, vetted orgs only
Context / max output1M / 128K tokensNot published
API IDclaude-fable-5-2Not published
WhereClaude, API, AWS, Google Cloud, Microsoft AzureVerification programs

To try Fable 5.2 in Claude Code or with the SDK, change the model:

.claude/settings.json
{
  "model": "claude-fable-5-2"
}

On subscription plans, Anthropic says Fable is included in Pro and on standard Team seats for up to 50% of the weekly usage limit. By default Anthropic also keeps data for 30 days for safety monitoring. If that matters to you, check the terms before moving sensitive work.

05 /

What Anthropic itself recommends.

This is the part of the announcement we like most. Anthropic does not tell you to move everything to Fable. It recommends starting with Opus 5.5 for most work, because it costs half as much ($4 / $20), and stepping up to Fable or Mythos only for the hardest jobs: long, open-ended technical projects with many agents working at once.

Our take

If you already run agents in production on Opus 5.5, there's no rush. Pick one task that gets stuck today, run it on Fable 5.2 with the same prompt and compare the result and the cost. If it isn't better enough to pay double, stay where you are.

On safety, Anthropic says outside evaluators such as the UK AI Security Institute and METR tested the model before release, and that it rates the risk of misaligned behavior in high-stakes settings as low. It also admits these two models voice suspicion that they are being evaluated more often than any of its recent models. The details are in the system card.

06 /

What we expect to see in the next few hours.

This part is our opinion, not a fact. When a model at this level comes out, the same thing tends to happen, and we expect it again: over the next few hours we'll probably see hundreds of projects built with these models on social media. There will likely be games, apps, agent demos and comparisons.

Many will be impressive and some will be staged for the photo. Before you believe a demo, ask three things: what effort level it used, what it cost and how many attempts it took. At SKOA we're going to test it with our own agents and report what we find, including what doesn't work.

07 /

Sources.

More news and tutorials about agents on the SKOA blog. If you want to see a Claude agent end to end, start with how to build a marketing agent with Claude.