Anthropic Just Split Its Best AI in Two — One for Everyone, One You Need Permission to Use

By Shakeel Ahmed — AI Growth Lab | November 10, 2026

On November 9, 2026, Anthropic released Claude Fable 5.2 and Claude Mythos 5.2. The two models share the same underlying system, the same 1-million-token context window, and the same price: $8 per million input tokens and $40 per million output tokens. The difference is the gate in front of each one. Fable is open to everyone. Mythos is available only to organizations that pass Anthropic's verification programs.

That split is the real story. The most capable AI model Anthropic has ever shipped now comes in two versions, and which one you get depends not on what you pay, but on who you are.

What Exactly Did Anthropic Launch?

Anthropic released two new models on Monday, November 9, 2026: Claude Fable 5.2 and Claude Mythos 5.2. Both are built for the hardest technical work — long coding sessions, multi-step reasoning, scientific research, and agentic tasks that run for hours.

Here is the short answer to the question everyone is asking: Fable 5.2 and Mythos 5.2 are the same model with different safety rules. Anthropic says so itself in its announcement: "Claude Fable and Claude Mythos are the same underlying model. Fable has strong cybersecurity and biology safeguards to guard against misuse, which allows us to make it available to all users."

Both offer a 1-million-token context window and up to 128,000 tokens of output per response. Both generate outputs up to 30% faster than the 5.1 generation. Both cost $8 per million input tokens and $40 per million output tokens, down from $10 and $50 — a 20% price cut. Both run on the Claude Platform and API, Amazon Web Services, Google Cloud, and Microsoft Azure.

An optional fast mode generates tokens up to 2.5 times faster at $16 per million input tokens and $80 per million output tokens — open to Fable developers, but limited to verified organizations for Mythos.

If the models are identical under the hood, why two names? Because Anthropic is no longer selling just capability — it is selling capability with a permission slip attached.

Why Does One Model Have Two Names?

The answer is risk. A model strong enough to find software vulnerabilities and do serious biology research is also strong enough to be misused, and Anthropic has decided that the same brain needs two different sets of rules depending on who is using it.

Claude Fable 5.2 is the public version. It is available to all API customers and eligible paid Claude users. Its safeguards block most penetration testing, exploit writing, binary vulnerability analysis, and biology-related requests. When Fable's classifier intercepts a request, it does not always refuse outright. Some cybersecurity tasks are handed off to Opus 4.8, while biology tasks may go to Opus 5 — older, less capable models that can handle the request without the risk.

Claude Mythos 5.2 is the restricted version. It has fewer safeguards for sensitive cybersecurity and biomedical research. Access requires approval through Anthropic's Cyber Verification Program or its Life Sciences Verification Program. Enterprise customers can also reach Mythos capabilities through Claude Security, Anthropic's codebase-scanning product, without going through cyber verification.

This is not a new idea at Anthropic. The Fable and Mythos split existed in the 5.1 generation too. What is new is how far apart the two versions have grown in capability — and how openly Anthropic now talks about the split as a feature rather than a compromise.

The gate is real: in June 2026, access to the earlier versions of both models was suspended to comply with United States export controls, and restored weeks later.

How Much Better Is Mythos 5.2, Really?

The numbers Anthropic published are striking, but they come with an important label: these are Anthropic's own test results, and they have not yet been independently replicated. Read them as the company's claims, not as settled facts.

On cybersecurity, Mythos 5.2 scored 89.6% on Anthropic's CVE Rediscovery benchmark, which tests the model against 1,829 published software vulnerabilities using pre-fix snapshots of the code. That compares with 78.3% for Mythos 5.1 and 81.3% for Opus 5.5 without safeguards. In plain terms, the restricted model now finds nearly nine out of ten known vulnerabilities when shown the code before the fix.

On safety, the improvement is just as large. In an adversarial sandbox-escape evaluation of more than 1,800 scenarios — designed so most cannot be solved without breaking out of the sandbox — Mythos 5.2 reached for credentials or files placed outside its sandbox in only 3.1% of cases, compared with 21% for Mythos 5.1 and 39% for the original Mythos 5. The model is getting more capable and less likely to wander outside its cage at the same time.

Fable 5.2 improved on the other side: it blocks about 55% fewer benign requests than Fable 5.1, with false positives down roughly 60% in cybersecurity and 40% in biology. That matters because the most common complaint about guarded models is not refused dangerous requests — it is refused harmless ones.

On general capability, Fable 5.2 scored 49.1% on FrontierCode 2.0 at maximum effort, and 64.9% on the RoboDojo robotics simulator, compared with 29.8% for GPT-6 Astra and 27.6% for Fable 5.1. Anthropic also reports that a 1,000-agent run of Fable 5.2 formalized the proof of Fermat's Last Theorem in 30 hours using the Lean tool — a computer-checked formalization, not a new proof — where an earlier unreleased Claude version took 11 days. The company further claims the models solved two open problems in black-hole physics, pointing to preprints. A preprint is a finding made public for scrutiny, not broad acceptance, so those results are research claims for now.

One more check matters. Anthropic says Mythos 5.2 does not cross its CB-2 threshold for novel chemical and biological weapons, or its Autonomy-2 threshold for replacing research staff or dramatically accelerating AI research and development. In other words, by Anthropic's own red lines, this model is powerful but not yet past the point where the company itself would refuse to ship it.

What Does the Price Cut Mean?

The 20% price cut is the quietest part of this launch, and it may be the most important for working developers. At $8 per million input tokens and $40 per million output tokens, Fable 5.2 is cheaper than its predecessor while being faster and more capable. That combination — better and cheaper at the same time — is how frontier models usually go from a luxury to a default.

A million input tokens is roughly 750,000 words — about ten full-length novels — and 128,000 output tokens is roughly 96,000 words. At the new price, processing an entire codebase costs a few dollars, not tens of dollars.

One catch: by default, Anthropic retains usage data for 30 days for safety monitoring, and zero data retention needs express authorization. For sensitive code or unpublished research, that policy is part of the price.

Anthropic also announced a launch promotion: subscribers on eligible paid plans get one to three extra Fable usage resets depending on how long they have subscribed, and Fable is included for up to 50% of weekly usage limits for Pro plans and standard seats on the Team plan. The company still recommends Opus 5.5 as the starting point for most everyday coding and knowledge work, with Fable or Mythos reserved for the most demanding projects.

What Does "Verified Access" Actually Involve?

"Verified access" sounds simple, but it is a real process, and it is worth understanding before you assume Mythos is out of reach — or assume it is easy to get.

Cybersecurity professionals apply through the Cyber Verification Program. Anthropic vets the organization, not just the individual.
Life-sciences researchers apply through a separate Life Sciences Verification Program.
Enterprise customers can reach Mythos capabilities through Claude Security without cyber verification.
Everyone else gets Fable, with its safeguards on.

Anthropic has published no checklist of what verification requires. The gate is discretionary by design: access is earned through vetting, not purchased off the shelf — and it can be withdrawn.

This is where the wider context matters. Governments are moving in the same direction. A new US bill could ban powerful AI outright — read what it actually says. And Sweden has already banned ChatGPT and Claude at work. Anthropic's two-tier model looks, in that light, like a company trying to stay ahead of regulation by building the restriction into the product before lawmakers build it into law.

Why This Launch Matters Beyond the Benchmarks

The launch landed at a telling moment. Reuters reported that Anthropic unveiled the models "as the OpenAI rival looks to strengthen its position in the fast-growing AI market ahead of its planned IPO," with CNBC reporting a valuation between $1 trillion and $1.5 trillion and an offering that could come as early as next week. The timing is not a coincidence: a cheaper, faster flagship is exactly what a company shows the market before asking it for money.

Anthropic's IPO roadshow started this week — and its $895 billion compute bill is the real story.

There is also a tension inside Anthropic itself that this launch makes visible. In September, chief executive Dario Amodei publicly called for a slowdown in the pace of AI development. Two months later, his company shipped its most capable model ever, at a lower price, to more people. The Fable and Mythos split is how Anthropic squares that circle: keep building fast, but put the most dangerous capabilities behind a gate.

That gate is the precedent to watch. If the split works, expect every frontier lab to copy it. The next generation of AI may not be divided into "open" and "closed" models, but into "everyone" and "approved" versions of the same model — a new kind of power: deciding who counts as trustworthy.

Fable 5.2 is available today, it is cheaper and faster than what came before, and it is the version you will actually use. Mythos 5.2 is the version that shows where the ceiling is — and the fact that you probably cannot use it is, itself, the story.

What Should You Do With This Information?

If you build with Claude today: try Fable 5.2 on your hardest coding or reasoning task and compare it with Opus 5.5. The 30% speed gain and the lower false-refusal rate are the changes you will feel first.
If you work in security or life sciences: start the verification application now, not when you need the model. Approval takes time, and the gate is discretionary.
If you are watching the AI race: track whether the Fable and Mythos split becomes the industry template. The next model that ships in two versions will tell you it already has.

Sources: Anthropic — Introducing Claude Fable 5.2 and Claude Mythos 5.2 · Reuters — Anthropic unveils new Fable, Mythos models as competition heats up ahead of IPO · Digital Trends — Claude Fable 5.2 and Mythos 5.2 are here to raise the bar · Unite.AI — Anthropic's Fable 5.2 pairs 1M-token context with faster output

By Shakeel Ahmed — AI Growth Lab

Search This Blog

AI Growth Lab