One-sentence summary
On November 9, 2026, Anthropic released Claude Fable 5.2 and Claude Mythos 5.2. Both are the same underlying model: Fable 5.2 ships with cybersecurity and biology safeguards and is available to all users, while Mythos 5.2 omits those limits and is available only to vetted cybersecurity and life-science organizations. Output is up to 30% faster than version 5.1, at $8 per million input tokens and $40 per million output tokens.
Event overview
On November 9, 2026, Anthropic published Claude Fable 5.2 and Claude Mythos 5.2 on its official site. The announcement lists three core changes: improvements in coding, reasoning and vision; output up to 30% faster than Claude Fable 5.1 and Claude Mythos 5.1; and a price cut to $8 per million input tokens and $40 per million output tokens.
The two models share one underlying model and differ in safeguards. Fable 5.2 carries cybersecurity and biology protections, which is why it can be released to all users. Mythos 5.2 does not carry those protections and is available only through Anthropic's Life Sciences Verification Program and Cyber Verification Program, plus Claude Security on Enterprise plans.
MacRumors added two details in its launch-day report: Fable 5.2 produces roughly 55% fewer blocks than Fable 5.1, and a fast mode can speed token generation by up to 2.5x. Chinese tech outlet Sina Tech reported that the previous generation was priced at $10 per million input tokens and $50 per million output tokens, making the new list price about 20% lower.
Key facts
| Item | What official and reported sources say |
|---|---|
| Release date | November 9, 2026 (Anthropic official site) |
| Models | Claude Fable 5.2, Claude Mythos 5.2 |
| Underlying relationship | Same underlying model, different safeguards |
| Fable 5.2 access | All users, with cybersecurity and biology safeguards |
| Mythos 5.2 access | Organizations vetted through Life Sciences / Cyber Verification programs, and Claude Security on Enterprise |
| Output speed | Up to 30% faster than 5.1 |
| Fast mode | Up to 2.5x faster token generation; $16 per million input tokens and $80 per million output tokens |
| Standard price | $8 per million input tokens, $40 per million output tokens |
| Versus previous generation | $10 / $50 previously, roughly a 20% cut (MacRumors, Sina Tech) |
| Safeguard difference | About 55% fewer blocks than 5.1 (MacRumors) |
| Subscription terms | Pro and Team standard seats can use up to half of their weekly quota for Fable this week with no extra charge, plus one Fable reset; Max, Ultra and Team premium seats receive one to three resets based on subscription length (reported) |
Capability claims: from formal proofs to electronic devices
Anthropic's announcement lists applications from prerelease testing and states that these results come from internal testing rather than third-party evaluation.
First, formal mathematics. Anthropic says the model formalized the proof of Fermat's Last Theorem in 30 hours. Sina Tech reported a comparison: under the same task configuration, Claude Opus 5.5 ran for 18 hours and completed only 26%. "Formalized" means translating a mathematical proof into a language a computer can check step by step; whether the proof is accepted by the mathematics community still requires independent proof assistants and peer review.
Second, electronic device design. A research post Anthropic published the same day says Fable 5.2, working with Diode Computers, designed three electronic devices: a speakerphone, a voice-assistant hub and a handheld virtual pet. The model laid out circuit boards, checked manufacturing constraints, modeled designs in CAD, and human engineers built prototypes that worked without any fixes to their circuitry. The reported benchmarks are 69% on PCB Gym versus 25% for the previous Fable 5, and 83.5% on BenchCAD 2.1, nearly double Fable 5.1.
Third, software migration and science. Anthropic says the model helped engineers complete otherwise-infeasible software migrations during prerelease testing, solved two open problems in black-hole physics, and made progress on robotics evaluations.
One model, two safeguard regimes
The Fable/Mythos split effectively separates capability from access. The underlying weights are the same; Fable routes high-risk requests through safeguards or hands them to other models, while Mythos gives vetted organizations the unrestricted capability.
Anthropic's announcement describes the routing rules: cybersecurity questions fall back to Claude Opus 4.8, and biology questions fall back to Claude Opus 5. That means Fable 5.2's scores in sensitive domains reflect the deliverable level after safeguards, not the model's ceiling.
The approach resolves two conflicting goals: releasing a highly capable model to as many paying users as possible, while reducing the risk that it is used for cyberattacks or bioweapon development. Running two configurations of one model is cheaper than training two independent models and lets the safety team maintain a single set of weights.
The cost is product complexity. Users face two names, two vetting processes, two price points and two capability boundaries within one model family, and procurement requires judging whether a use case falls into a restricted category.
Pricing and subscriptions: what a 20% cut means
At $8 per million input tokens and $40 per million output tokens, the new price is about 20% below the previous $10 / $50. Cutting prices while raising capability has become the norm in frontier-model competition in 2026: when Anthropic released Claude Opus 5.5 on September 22, it also cut the cost of typical workloads by 40% versus Opus 5.
The practical impact depends on workload shape. Output tokens cost five times as much as input tokens, so long answers, code generation and multi-step agent reasoning are driven mainly by the output side; input-side cuts help long-document and codebase-retrieval applications more directly.
The subscription arrangement looks more like a limited-time promotion. According to MacRumors and Sina Tech, Pro and Team standard seats can apply up to half of their weekly quota to Fable this week at no extra charge and receive one Fable reset; Max, Ultra and Team premium seats receive one to three resets depending on subscription length, valid through March 9, 2027. Anthropic's official announcement does not state these subscription details, so this site labels them as reported terms.
Impact and limits
First, model tiering becomes finer. Anthropic's guidance is to start most coding and knowledge tasks with Claude Opus 5.5, switch up to Fable or Mythos for harder problems, and step down to Sonnet 5.5 or Haiku 5.5 for lower cost. That "daily driver plus heavy-lift model" combination is replacing the practice of buying one model for every task.
Second, safeguards become a priced product attribute. Fable 5.2 produces roughly 55% fewer blocks than 5.1 while, according to Anthropic, keeping protections as robust as 5.1's. For developers, the false-positive rate directly affects usability; for regulators, the presence of safeguards determines whether a model can enter regulated environments.
Third, benchmark claims need qualification. The capability figures in the announcement come from Anthropic's internal tests and in-house benchmarks, and no third-party replication has appeared. The figures Sina Tech cited — 45.3% on coding at High effort versus 45.5% for Opus 5.5 at Max effort — are vendor claims relayed by media, not independent verification.
The limits are explicit: Anthropic did not disclose how many organizations can access Mythos 5.2, Fable 5.2's traffic capacity or rate limits, or the full verification results behind the 30-hour formal proof. This site previously covered the Claude Opus 5.5 launch (Claude Opus 5.5 launch); this article does not repeat Opus 5.5 benchmarks or pricing.