Quick Answer
Claude Fable 5 and Claude Mythos 5 are not a conventional fast-versus-smart model pair. Anthropic says they use the same underlying model and share specifications and published pricing.
The decision is primarily about safeguards and access:
| Dimension | Claude Fable 5 | Claude Mythos 5 |
|---|---|---|
| Intended access | Broad, generally available | Approved organizations only |
| Safeguards | Conservative classifiers for general use | Safeguards lifted in approved domains |
| Safeguard result | Relevant requests can fall back to Opus 4.8 | Governed by trusted-access program |
| Self-serve sign-up | Available through normal provider products and API, subject to entitlement | None |
| API ID | claude-fable-5 | claude-mythos-5 after approval |
| Context | 1 million tokens | Shares Fable specifications |
| Max synchronous output | 128,000 tokens | Shares Fable specifications |
| Published API price | $10 input / $50 output per million tokens | Same published price for approved use |
| Flowith status | Verify separately | Verify separately; provider access is restricted |
Choose Fable 5 for general agentic, coding, vision, research, and knowledge work when its safeguard behavior is acceptable. Pursue Mythos only when an eligible organization has a defensive use case, governance capacity, and explicit provider approval.
Use the Claude Fable 5 model page and Claude Mythos 5 model page for the current entity records. For subscription details, use the Fable 5 plans and credits guide.
Source check: Anthropic’s launch, redeployment, current model overview, plan article, and covered-model retention policy were checked July 30, 2026.
The Shared Model
Anthropic launched Fable 5 as a Mythos-class model made safe for general use and Mythos 5 as the same underlying model with safeguards lifted in some areas.
Current model documentation says Mythos shares Fable’s:
- specifications;
- pricing;
- text and image inputs;
- text output;
- vision and multilingual capability;
- 1 million token context window;
- 128,000-token maximum output in the synchronous Messages API.
Do not attribute a performance difference to the names alone. A user can still observe different behavior because safeguards, fallback, access environment, tools, system instructions, quotas, and governance differ.
The Fable Safeguard Profile
Anthropic describes a defense-in-depth approach for Fable 5. The visible runtime behavior centers on classifiers covering:
- cybersecurity;
- biology and chemistry;
- distillation attempts.
When a relevant classifier triggers, Anthropic says the request can be routed to Claude Opus 4.8 and the user is notified.
This matters in evaluation:
- the answer may come from a different model;
- benign work can produce false positives;
- agent state and tool use need fallback testing;
- repeated blocked prompts are not evidence that Mythos should be self-provisioned;
- product logs should record the model or fallback shown by the provider when available.
Anthropic reported that more than 95% of early Fable sessions involved no fallback. Treat this as launch evidence, not a guarantee for a specialized workload.
The Mythos Trusted-Access Profile
Mythos 5 is designed for approved organizations needing capabilities that Fable’s general safeguards restrict.
Anthropic’s current sources describe:
- Project Glasswing access for defensive cybersecurity;
- restoration to a set of approved US organizations;
- plans to expand to broader domestic and international partners;
- a separate small biology program that can remove biology and chemistry safeguards while retaining cyber safeguards;
- invitation-only access with no self-serve sign-up.
The phrase “safeguards lifted” does not mean unrestricted use. Access remains bounded by:
- organization approval;
- program purpose;
- provider and government coordination where applicable;
- acceptable-use terms;
- monitoring and review;
- security and incident-response controls;
- human oversight;
- retention requirements.
Access Decision Tree
Choose Fable 5 When
- the user or organization needs a generally available highest-capability Claude model;
- normal Claude products, API, or supported cloud access fits;
- conservative fallback behavior can be tested and accepted;
- the current plan or usage-credit cost is approved;
- covered-model retention is compatible with the data policy.
Consider Mythos 5 Only When
- the organization is eligible for Project Glasswing or another trusted-access program;
- the use case is defensive and within the approved domain;
- provider approval and provisioning are complete;
- governance, logging, human review, and incident response are documented;
- the organization accepts the covered-model data terms;
- no public or lower-risk model satisfies the requirement.
Choose Another Model When
- Fable’s safeguards cause unacceptable false positives;
- Fable’s price or latency does not fit;
- the workload does not need Fable-level capability;
- Mythos access is unavailable or inappropriate;
- retention conflicts with organizational requirements.
Claude Opus 5, Sonnet 5, or another supported model may provide a better price, latency, governance, or access fit.
API and Provider Access
Fable 5
Anthropic documents general first-party API availability under:
claude-fable-5
Cloud platform availability still depends on current region, platform ID, quota, and contract.
Mythos 5
Anthropic names:
claude-mythos-5
The ID is not a self-serve credential. Approved customers should work with Anthropic, AWS, or Google Cloud account teams and use the exact provisioned endpoint.
Do not:
- probe restricted IDs from an unapproved account;
- infer access from a documentation table;
- copy Fable credentials or entitlements;
- market Mythos as a paid consumer upgrade;
- imply Flowith access from a provider model ID.
Price Is Not the Access Gate
Anthropic published the same $10 per million input tokens and $50 per million output tokens for both models.
That does not make Mythos purchasable on demand. Mythos approval and program scope are separate from consumption pricing.
For Fable, price also appears through different billing surfaces:
- standard API consumption;
- usage credits in Claude products;
- cloud platform billing;
- enterprise contracts.
Verify prompt caching, batch pricing, tools, long outputs, retries, and platform markup before estimating a workload.
Covered-Model Retention
Fable 5 and Mythos 5 are covered models under Anthropic’s policy.
The policy change specifically affects organizations that otherwise have zero-data-retention arrangements on Claude Console, Claude Enterprise Code, or supported cloud platforms. Those organizations must permit 30-day retention of prompts and outputs for covered models.
Anthropic states:
- the data supports safety work;
- it is not used to train new Claude models;
- default human access is blocked;
- flagged review uses a controlled path;
- access is logged;
- deletion occurs after 30 days in almost all cases;
- safety or legal exceptions can extend retention.
Consumer surfaces were already retained under separate terms. Evaluate the exact product and contract rather than promising a universal 30-day deletion rule.
Flowith Availability Is Independent
Neither general Fable availability nor approved Mythos access establishes:
- a Flowith mapping;
- plan entitlement;
- model routing;
- provider fallback visibility;
- context and output limits;
- pricing;
- retention or data handling.
Use the live Flowith workspace and first-party Flowith documentation. Google Project Glasswing, Claude Cowork, and Flowith are distinct products and programs.
Evaluation Checklist
Test Fable before pursuing restricted access:
- representative allowed prompts;
- cyber, bio, or chemistry-adjacent benign prompts likely to trigger safeguards;
- user-visible fallback notification;
- model identity and audit logging;
- tool execution after fallback;
- latency and output consistency;
- plan or API cost;
- retention and data classification;
- human review of high-impact outputs;
- a lower-risk fallback model.
If Fable fails the acceptance bar, document the exact blocked capability and business need. That evidence may support a provider conversation; it does not create Mythos eligibility.
Frequently Asked Questions
Is Mythos 5 more capable than Fable 5?
Anthropic describes the same underlying model. Mythos can expose capabilities that Fable’s safeguards block or route to Opus 4.8, so the practical difference is safeguard-mediated access rather than a published base-model specification gap.
Can I pay extra to unlock Mythos 5?
No self-serve path is documented. Approved organizations must enter Project Glasswing or another trusted-access program.
Does Fable 5 always answer with Fable 5?
No. Anthropic says classifiers can route relevant requests to Opus 4.8 and notify the user.
Does Mythos 5 have different pricing?
Anthropic published the same base token pricing as Fable 5, but approved program and cloud contracts can add separate terms.