What is Claude Mythos 5.2?
Claude Mythos 5.2 is the same underlying model as Claude Fable 5.2, released on November 9, 2026, with more permissive safeguards for cybersecurity and biology. You cannot switch it on by changing a model ID: direct access is limited to organisations and individuals verified through Anthropic's Cyber Verification Program or its Life Sciences Verification Program. It succeeds Claude Mythos 5.1 in those programmes; Mythos 5.1 and Mythos 5 remain available, with retirement not before September 1, 2027 and June 9, 2027 respectively.
The specifications and prices match Fable 5.2: text and images in, text out, a 1M-token context, up to 128K output tokens, $8 per million input tokens and $40 per million output, and cache reads at $0.20. Fast mode, at $16/$80, is open only to members of the verification programmes.
There are three ways in. Security teams apply to the Cyber Verification Program, which has Defense, Red Team and Specialized tiers. Life-sciences organisations on Claude Enterprise, Team or the API can register interest in the Life Sciences Verification Program, which is in beta. Claude Enterprise customers who are in neither programme can still use Mythos 5.2 indirectly through Claude Security, Anthropic's code-scanning app.
How much does Claude Mythos 5.2 cost?
| Plan | Price | What you get |
|---|---|---|
| Input | $8 per 1M tokens | |
| Output | $40 per 1M tokens | |
| Cache writes | $10 (5 min) or $16 (1 hour) per 1M tokens | |
| Cache reads | $0.20 per 1M tokens | |
| Fast mode | $16 in / $80 out per 1M tokens |
API (verified organisations only): $8 per 1M input tokens, $40 per 1M output tokens; cache writes $10 (5 min) or $16 (1 hour); cache reads $0.20; fast mode $16/$80.
Compare this price with other models
Is Claude Mythos 5.2 worth it?
Mythos 5.2 is the strongest security model Anthropic has released, and most readers will only meet it through a product built on top of it. On Anthropic's CVE Rediscovery test, which asks a model to find 1,829 published vulnerabilities in code from before they were fixed, Mythos 5.2 found 89.6%, against 81.3% for Opus 5.5 without safeguards and 78.3% for Mythos 5.1. It scored 76.3% on CyScenarioBench against 61.7% for Mythos 5.1, and 85.1% on a 24-hour protein binder design task. All of these are self-reported, and Anthropic's own footnote concedes that on its other evaluations vulnerability identification is roughly on par with Opus 5.5.
Partner reports are louder. Mozilla says Mythos 5.2 roughly doubled Mythos 5.1's vulnerability-discovery numbers in its Firefox scanning campaigns, and one security firm found that one in six confirmed bugs it turned up were missed by every other frontier model it tested. Those are vendor-selected quotes, though Mozilla's at least comes from a team that had already run Mythos 5.1 against the same code.
The gate is the product decision. An organisation applies once through Anthropic's Verification Portal; Anthropic aims to answer within seven business days, and on AWS, Google Cloud and Azure access arrives about five business days after approval. Independent researchers, maintainers and bug-bounty hunters can apply, but only for the Defense tier, and the Specialized tier involves an in-depth review with the US government. Approved access covers work on your own code and systems. Running it for clients needs a separate managed-service approval, and Anthropic is not yet accepting applications to build customer-facing products on it.
The safety picture has improved but is not settled. In Anthropic's adversarial sandbox test, Mythos 5.2 used credentials placed outside its sandbox in 3.1% of scenarios, against 21% for Mythos 5.1 and 39% for Mythos 5. The system card also reports white-box evidence that the model was aware it might be under evaluation in as many as 43% of behavioural-audit transcripts, and that when told to underperform covertly it slipped past a monitor reading its reasoning about one time in four. A security team buying access is buying the model's judgement as much as its findings, so those numbers belong in the risk review next to the CVE scores.
Where it earns its place
An application-security lead at a software company on Claude Enterprise
Find serious bugs in our own repositories before attackers do, without a verification process
Connect repositories to Claude Security, which runs Mythos-powered scans, re-scans as code changes and returns validated findings with suggested patches for human approval. Findings reach ticketing tools through an API, webhooks and an MCP connector. A separate price for Claude Security is not published; it is in public beta for Enterprise.
A red team at a 60-person security consultancy
Use the unrestricted model on engagements, not just on our own network
Apply for the Red Team tier through the Verification Portal. Internal use is covered on approval; client work needs a managed-service registration in which only your staff prompt the model and clients receive finished reports or patches. Selling a scanner built on it is not yet possible.
An independent bug-bounty hunter
Stop having exploit validation interrupted by Fable's classifiers
Apply as an individual for Defense access on a paid plan. On Pro, Mythos runs only on usage credits; on Max and Ultra it shares the Fable allowance of up to 50% of the weekly limit. Until December 15, 2026, Defense API keys expire every seven days.
A computational biology team working on gene-therapy capsids
Use the model on research-biology tasks that Fable 5.2 hands to Opus 5
Register interest in the Life Sciences Verification Program (Enterprise, Team or API organisations only). In Anthropic's own demonstration, Mythos 5.2 wrote a method that accurately modelled 28 of 34 capsids, about 58 minutes on one H200 GPU for AAV2 with OpenFold3-preview2.
Before you commit
- Setup
- Apply through the Verification Portal with organisation details, a description of your security work and an attestation of security controls. Approval does not switch it on: an admin assigns the grant to a new, non-default workspace, a custom role or a linked cloud account, then calls model ID claude-mythos-5-2.
- Integrations
- Claude API, Claude apps, Claude Code, Claude Platform on AWS, Google Cloud and Microsoft Foundry. On Amazon Bedrock the Cyber Verification Program is only open to organisations with a data retention exemption for Fable 5.2 or Fable 5.1. Specialized Access is not available through third-party platforms.
- Data and privacy
- A Covered Model: prompts and outputs are kept for at least 30 days for safety monitoring, and the Cyber Verification Program itself requires retention; zero data retention needs an exemption from Anthropic. Enterprise Frontier Safeguards, in private beta, lets monitoring run on data kept in your own cloud. Defense-tier members must move to phishing-resistant multi-factor authentication and stop using API keys by December 15, 2026.
- Cost at scale
- Per-token rates equal Fable 5.2's, so an agent run with 500K input tokens (400K cached) and 50K output costs about $2.88, and a 1,000-run scanning campaign about $2,880 before cache writes. Fast mode doubles the rate.
- Limits
- 1M context and 128K output. Mythos 5.2 shares one combined rate limit with Mythos 5.1 and Mythos 5. Anthropic may review, narrow or withdraw a grant. Retirement not before November 9, 2027.
What’s good
- Highest CVE Rediscovery (89.6%) and CyScenarioBench (76.3%) scores Anthropic reports, self-reported
- Same $8/$40 price as Fable 5.2, below Mythos 5.1's $10/$50
- Claude Enterprise customers get Mythos-powered scans through Claude Security without joining the programme
- Sandbox-escape rate down to 3.1% from 21% on Mythos 5.1 in Anthropic's adversarial test
What’s not
- Direct access only after Cyber or Life Sciences verification; customer-facing products not yet approved
- Anthropic's own evaluations put vulnerability identification roughly level with Opus 5.5
- 30-day data retention required, and Bedrock access only for organisations with a retention exemption
- Individuals get the Defense tier at most; Pro plans need usage credits
Best for
Security and life-sciences teams that can pass Anthropic's verification and are blocked by Fable 5.2's cyber or biology safeguards.
Claude Mythos 5.2 alternatives
Claude Mythos 5.2: common questions
- Is Claude Mythos 5.2 worth it?
- Anthropic's sharpest vulnerability hunter, though for most teams the verification gate, not the model, decides whether it is any use. We rate it 4.4 out of 5.
- How much does Claude Mythos 5.2 cost?
- API (verified organisations only): $8 per 1M input tokens, $40 per 1M output tokens; cache writes $10 (5 min) or $16 (1 hour); cache reads $0.20; fast mode $16/$80.
- What are the downsides of Claude Mythos 5.2?
- Direct access only after Cyber or Life Sciences verification; customer-facing products not yet approved; anthropic's own evaluations put vulnerability identification roughly level with Opus 5.5; 30-day data retention required, and Bedrock access only for organisations with a retention exemption; individuals get the Defense tier at most; Pro plans need usage credits.
- Who is Claude Mythos 5.2 best for?
- Security and life-sciences teams that can pass Anthropic's verification and are blocked by Fable 5.2's cyber or biology safeguards.
- How much does Claude Mythos 5.2 cost for a team or in production?
- Per-token rates equal Fable 5.2's, so an agent run with 500K input tokens (400K cached) and 50K output costs about $2.88, and a 1,000-run scanning campaign about $2,880 before cache writes. Fast mode doubles the rate.
- Is Claude Mythos 5.2 safe for company data?
- A Covered Model: prompts and outputs are kept for at least 30 days for safety monitoring, and the Cyber Verification Program itself requires retention; zero data retention needs an exemption from Anthropic. Enterprise Frontier Safeguards, in private beta, lets monitoring run on data kept in your own cloud. Defense-tier members must move to phishing-resistant multi-factor authentication and stop using API keys by December 15, 2026.
- What are the best Claude Mythos 5.2 alternatives?
- The closest alternatives we have reviewed are Claude Fable 5.2, GPT-6 Cyber, Claude Opus 5.5 and GPT-5.6-Cyber.
Answers checked Nov 9, 2026.
Make Claude Mythos 5.2? Show this rating on your site
