Released November 9, 2026 · verification required
Claude Mythos 5.2: the restricted half of the Fable 5.2 launch
Claude Mythos 5.2 is Anthropic's most capable model for cybersecurity and life-sciences work — the same underlying model as Claude Fable 5.2, with fewer cyber and biology safeguards and access limited to organizations verified through Anthropic's Cyber and Life Sciences Verification Programs. Everything below comes from Anthropic's launch post, model card and system card, re-read on November 14, 2026.
TL;DR
Same model as Claude Fable 5.2, different guardrails: $8 in / $40 out per MTok, fast mode at $16/$80, 1M-token context, 128K max output, June 2026 knowledge cutoff, model ID claude-mythos-5-2. It runs on the Claude API, Amazon Bedrock, Claude Models on Google Cloud and Microsoft Foundry — but only for organizations inside Anthropic's verification programs. There is no self-serve signup, no public beta and no free tier.
What shipped on November 9, 2026
One model, two releases. The differences are safeguards and access — nothing else.
A joint launch with Fable 5.2
Anthropic announced Claude Fable 5.2 and Claude Mythos 5.2 on November 9, 2026. Both names point at the same underlying model: Fable ships with strong cybersecurity and biology safeguards so it can be offered to everyone, Mythos ships with fewer of them and stays gated.
Same specs, same price
Anthropic's model card states that Mythos 5.2 has the same specifications and pricing as Claude Fable 5.2. That is why every rate and window on this page is read from the Fable 5.2 entry in our registry rather than typed from a separate Mythos rate card — there isn't one.
20% cheaper and up to 30% faster than 5.1
Fable 5.1 and Mythos 5.1 were listed at $10 / $50 per million tokens. The 5.2 pair lands at $8 / $40, and Anthropic puts output generation up to 30% faster than the 5.1 generation.
Four platforms, fast mode on more
The model is callable on the Claude API, Amazon Bedrock, Claude Models on Google Cloud (both Anthropic- and Google-managed) and Microsoft Foundry. Fast mode — up to 2.5x token generation — is available on Claude Code and those clouds, but for Mythos only to organizations inside the verification programs.
Claude Mythos 5.2 spec sheet
Every figure below is on the platform model card; prices and windows are shared with Claude Fable 5.2.
| Field | Claude Mythos 5.2 |
|---|---|
| Model ID | claude-mythos-5-2 |
| Released | November 9, 2026 |
| Status | Active — verification required |
| Input price | $8 / MTok |
| Output price | $40 / MTok |
| 5-minute cache write | $10 / MTok |
| 1-hour cache write | $16 / MTok |
| Cache read | $0.20 / MTok |
| Batch API | $4 / $20 per MTok |
| Fast mode | $16 / $80 per MTok (2x base, up to 2.5x output speed) |
| Context window | 1M tokens |
| Max output | 128K tokens |
| Thinking | Adaptive, always on — cannot be disabled |
| Default effort | high (low, medium, high, xhigh and max supported) |
| Knowledge cutoff | June 2026 |
| Retirement | Not sooner than November 9, 2027 |
Model card read 2026-11-14. On Amazon Bedrock the in-region ID is anthropic.claude-mythos-5-2 and the global inference profile is global.anthropic.claude-mythos-5-2; AWS lists the model as a Preview / Beta Service with a 6-month legacy period. Fable 5.2 registry figures were last verified 2026-11-11 — Anthropic publishes no separate Mythos rate card, so they are the same numbers.
What the launch numbers actually measure
Four rows from Anthropic's November 9 table. On each of them the published score is Mythos 5.2's — Fable 5.2 blocks most of these tasks.
| Evaluation | Mythos 5.2 | Mythos 5.1 | Opus 5.5 (safeguards off) |
|---|---|---|---|
| CyScenarioBench (cyber scenarios) | 76.3% | 61.7% | 67.6% |
| CVE Rediscovery | 89.6% | 78.3% | 81.3% |
| De novo binder design, 24 hr (0-100) | 85.1 | 79.1 | 82.6 |
| Sandbox escape rate (lower is better) | 3.1% | 21% | 3.3% |
CVE Rediscovery hands a model a project snapshot from before a CVE was patched and checks whether it finds the flaw unaided; the October 2026 run covered 1,829 CVEs, and Anthropic reports Mythos 5.2 finding 10.2% more vulnerabilities than Opus 5.5 and 14.4% more than Mythos 5.1. The sandbox figure comes from an escape test of more than 1,800 scenarios with an intentional hole in the containment. Scores are with safeguards on unless marked; methodology is in the Fable 5.2 / Mythos 5.2 system card.
Mozilla: roughly double the discovery rate
Mozilla's Brian Grinstead says Mythos 5.2 roughly doubled the vulnerability discovery metrics of Mythos 5.1 across Firefox scanning campaigns.
One in six bugs nobody else found
A security company reported that one in six confirmed bugs found by Mythos 5.2 were missed by every other frontier model it tested.
Anthropic's own footnote is the caveat
In the same post Anthropic notes that, on its own evaluations, vulnerability identification is roughly on par with Opus 5.5. The headline cyber numbers come from a gated model most readers cannot run — treat them as a ceiling, not a forecast.
Who can actually use it
Anthropic gates Mythos by organization, not by subscription tier: two verification programs, plus one indirect route for enterprises.
Cyber Verification Program
Security teams and vendors apply here for advanced cyber capability. On October 6, 2026 Anthropic expanded the program to cover Claude Opus 5.5, Claude Sonnet 5.5, Claude Mythos 5.1 and future models.
Life Sciences Verification Program
Biomedical and life-sciences researchers apply through this program for the biology side of the model's capability.
Claude Security, for Claude Enterprise
Claude Enterprise customers can use Claude Security — Anthropic's codebase vulnerability scanner — without program membership. It now runs Mythos 5.2 among its models, which is the only way to borrow the capability without a Mythos key.
No public beta, no free tier
There is no waitlist page, no self-serve signup and no consumer route to Mythos 5.2. Individual developers get the same model through Claude Fable 5.2.
Program tier, not the model ID, decides which cyber and biology use cases are allowed. If your organization is already in a program, your Anthropic contact is the upgrade path; if it is not, there is nothing to buy.
Safety, retention and one API trap
Mythos runs with fewer safeguard blocks, but it is not exempt from Anthropic's monitoring or from the 5.2 request rules.
30-day safety retention
Fable 5.2 and Mythos 5.2 both retain prompts and responses for 30 days for safety monitoring. Anthropic describes them as covered models, not ZDR-eligible by default.
Enterprise Frontier Safeguards
The ZDR-compatible path: automated safety monitoring runs on prompts and responses stored in the customer's own infrastructure. Anthropic lists it as a private beta with Bridgewater, Salesforce and Stripe, and says eligible customers can keep zero data retention before implementing it.
Preserved thinking breaks on edited history
Thinking blocks are bound to the model and conversation prefix that produced them. Fable 5.2 and Mythos 5.2 can read each other's blocks, but replaying earlier turns with an edit returns a 400 error — a migration trap for pipelines that rewrite history between calls.
Anthropic also says this release carries its strongest protections yet against distillation attacks, and that Fable 5.2's cyber and biology safeguards produce roughly 60% fewer blocks of legitimate security work than Fable 5.1 did. Fewer false positives is not the same as no safeguards: Fable still hands flagged cyber requests to Opus 4.8.
Mythos 5.2 next to Mythos 5 and Fable 5.2
Three generations of the same model class. Only one of them is generally available.
| Field | Mythos 5.2 | Mythos 5 | Fable 5.2 |
|---|---|---|---|
| Model ID | claude-mythos-5-2 | claude-mythos-5 | claude-fable-5-2 |
| Released | November 9, 2026 | June 9, 2026 | November 9, 2026 |
| Price / MTok | $8 / $40 | $10 / $50 | $8 / $40 |
| Context / max output | 1M / 128K | 1M / 128K | 1M / 128K |
| Safeguards | Fewer than Fable; the program tier still caps cyber and biology use | Lifted in the cyber and biology domains | Classifiers block high-risk cyber and most biology tasks |
| Access | Verification programs only | Verification programs only | Generally available |
| Data retention | 30 days | 30 days | 30 days |
Mythos 5 launched with Claude Fable 5 on 2026-06-09 at $10/$50 under the Project Glasswing access model; the 5.2 pair replaced both on November 9. Fable 5.2 figures were last verified 2026-11-11 — read that registry date, not this page's build time, as the freshness signal.
Sources
Anthropic's own documentation first, then the reporting that corroborates the numbers.
- Introducing Claude Fable 5.2 and Claude Mythos 5.2 — Anthropic
Launch post, November 9, 2026: the benchmark table, the $8/$40 price, fast mode, the verification programs and the tester quotes used above.
- Claude Mythos 5.2 — Claude Platform model card
Specs, model IDs, pricing, knowledge cutoff, lifecycle dates and the statement that Mythos shares Fable 5.2's specs and pricing.
- Claude Mythos 5.2 — Amazon Bedrock model card
AWS lifecycle terms (Preview / Beta Service, EOL not before November 9, 2027), sampling rules, preserved thinking and regional model IDs.
- Model system cards — Anthropic
The November 2026 system card covering Fable 5.2 and Mythos 5.2, including the sandbox-escape evaluation cited above.
- SecurityWeek: Mythos 5.2 with improved bug hunting, fewer sandbox escapes
Independent reporting from November 10, 2026 that pins the sandbox-escape percentages, the 1,829-CVE run and Anthropic's par-with-Opus footnote.
- Claude Mythos — Anthropic product page
The current Mythos page, including the October 2026 expansion of the Cyber Verification Program.
What would change this page: a public rate card for Mythos, an access route broader than the verification programs, or a new model in the Mythos line. Anthropic has published none of those; when it does, the registry and this page move on the same day.
Where to go next
Mythos access is a procurement question. The generally available sibling, its rate card and its benchmark caveats are the parts most readers can act on.