Last updated: July 18, 2026 · Originally published March 28, 2026 · Availability, model IDs, pricing and platform guidance verified against current Anthropic and Claude Platform sources
Claude Mythos 5 is Anthropic’s limited-availability model for advanced cybersecurity and biology research. It is currently available by invitation to approved Project Glasswing customers under the API model ID claude-mythos-5; there is no self-serve sign-up. Claude Fable 5 uses the same underlying model with additional safeguards and has been generally available since June 9, 2026. Teams without Mythos access can use Fable 5 for the same core capabilities, while Claude Opus 4.8 remains a broadly available option for complex agentic coding and enterprise work.
claude-mythos-5claude-fable-5What is Claude Mythos 5?
Claude Mythos 5 is the current production name for Anthropic’s most capable model in trusted-access cybersecurity and biology programs. Anthropic describes it as the successor to Claude Mythos Preview and says it shares Claude Fable 5’s specifications, pricing and underlying capabilities.
The key difference is the access and safeguard model. Mythos 5 is offered to a small, growing group of vetted customers for high-value dual-use research. Fable 5 packages the same underlying model with strong classifiers intended to make Mythos-level knowledge work, coding and agentic capability suitable for broad release.
That distinction makes “Claude Mythos” a current product-status query, not a future-model rumor. The relevant questions are whether an organization qualifies for Project Glasswing, whether its workload needs unrestricted cyber or biology capability, and whether it can accept the model’s data-retention terms.
Claude Mythos 5 current availability and access
Mythos 5 is active but not generally available. Anthropic’s model overview calls it limited availability, and its current product documentation says access is invitation-only with no self-serve sign-up. Approved customers should work through their Anthropic, AWS or Google Cloud account team.
| Access question | Current answer |
|---|---|
| Can any developer select Mythos 5 in the API Console? | No. The model is limited to approved Project Glasswing customers. |
| Is there a public application form? | No self-serve sign-up is listed. Anthropic directs organizations to an Anthropic, AWS or Google Cloud account team. |
| What is the API ID? | claude-mythos-5. The earlier trusted-access model is claude-mythos-preview. |
| What should a team use without access? | Anthropic recommends Claude Fable 5 for the same capabilities under broad availability and stronger safeguards. |
| What data policy applies? | Mythos 5 and Fable 5 require 30-day data retention and are not available with zero data retention. |
If your account has not been explicitly approved for Mythos 5, do not hard-code it as a production dependency. Use the generally available Claude Fable 5 path for the highest broadly released capability or Opus 4.8 where its capability, cost and established availability better fit the workload.
Timeline: from Mythos Preview to Mythos 5
- April 7, 2026Anthropic announces Project Glasswing. Launch partners and additional critical-software organizations begin using Claude Mythos Preview for defensive cybersecurity.
- May 28, 2026The Opus 4.8 announcement describes Mythos Preview as a higher-capability class used by a small number of organizations through Glasswing, while Opus 4.8 becomes the broadly available flagship for complex work.
- June 9, 2026Anthropic launches Claude Fable 5 in general availability and Claude Mythos 5 in limited availability. Mythos 5 succeeds Mythos Preview; Fable 5 brings the same core capability to broad use with additional safeguards.
- July 18, 2026Claude Platform documentation lists Fable 5 as generally available and Mythos 5 as invitation-only for approved Project Glasswing customers. Both have current API IDs, specifications and pricing.
Claude Mythos 5 vs Claude Fable 5
Fable 5 and Mythos 5 are not successive capability tiers. Anthropic says they share the same underlying model, specifications and base token pricing. Fable changes how risky cybersecurity and biology requests are handled so the model can be offered broadly.
| Dimension | Claude Fable 5 | Claude Mythos 5 |
|---|---|---|
| Availability | Generally available from June 9, 2026. | Limited availability for approved Project Glasswing customers. |
| Claude API ID | claude-fable-5 | claude-mythos-5 |
| Underlying capability | Same underlying model and general capabilities as Mythos 5. | Same underlying model, focused on trusted cyber and biology access. |
| Safety classifiers | Includes classifiers that can decline or reroute risky requests. | Does not include the Fable classifiers; access is controlled at the customer-program level. |
| Best fit | Long-running agents, ambitious coding and knowledge work under broad availability. | Approved defensive cybersecurity and biology research requiring the trusted-access capability. |
| Base token pricing | $10 per million input tokens and $50 per million output tokens. | $10 per million input tokens and $50 per million output tokens. |
| Context / maximum output | 1M context window / up to 128k output tokens. | 1M context window / up to 128k output tokens. |
| Data retention | 30 days; no zero-data-retention option. | 30 days; no zero-data-retention option. |
For Fable integrations, a classifier decline is returned as a successful HTTP 200 response with stop_reason: "refusal". Applications should handle that state and can configure Opus 4.8 as a fallback. This classifier behavior applies to Fable 5, not Mythos 5.
Claude Mythos 5 vs Claude Opus 4.8
Mythos 5 is the higher-capability trusted-access option for cybersecurity and biology, while Opus 4.8 is a broadly available model built for complex agentic coding and enterprise work. They are separate model IDs with different prices and access requirements.
| Question | Claude Mythos 5 | Claude Opus 4.8 |
|---|---|---|
| Who can use it? | Approved Project Glasswing customers only. | Generally available across Claude’s supported first-party and cloud surfaces. |
| Model ID | claude-mythos-5 | claude-opus-4-8 |
| Primary positioning | Advanced cybersecurity and biology research. | Complex agentic coding and enterprise work. |
| Base token pricing | $10 input / $50 output per million tokens. | $5 input / $25 output per million tokens. |
| Default recommendation | Only when the organization is approved and the workload needs Mythos-specific trusted access. | A practical default for complex production workloads that do not require Mythos access; see the current Claude Opus 4.8 status and model ID. |
Capabilities, specifications and API behavior
Anthropic positions Mythos 5 as its most capable model for cybersecurity and biology research. The current model page also highlights healthcare, while the shared Fable capability is aimed at long-running agents, coding, knowledge work, vision and computer use.
| Technical field | Claude Mythos 5 current value | Developer implication |
|---|---|---|
| Context window | 1 million tokens by default. | Design long-context prompts carefully; context size does not remove the need for retrieval and compaction strategy. |
| Maximum output | Up to 128k tokens per synchronous request. | Set practical output budgets rather than defaulting to the maximum. |
| Thinking mode | Adaptive thinking is always on. | Use the effort parameter to control depth; disabling thinking is not supported. |
| Thinking output | Raw chain of thought is never returned. | Use summarized or omitted thinking display and preserve returned thinking blocks correctly in multi-turn work. |
| Selected features | Effort, task budgets beta, memory, code execution, programmatic tool calling, context editing, compaction and vision. | Verify the current feature page and beta headers before deployment. |
| Model versioning | claude-mythos-5 is a pinned dateless model ID. | Dateless IDs from the Claude 4.6 generation onward are fixed snapshots, not evergreen aliases. |
Project Glasswing and why access is limited
Project Glasswing is Anthropic’s collaboration with major technology companies, critical-infrastructure organizations and open-source maintainers. Its purpose is to apply frontier cyber capability to defensive work: finding vulnerabilities, helping maintainers patch them and learning how to operate powerful models safely.
Mythos Preview was the original model used in the program. Mythos 5 is its current successor. Anthropic limits access because the same skills that help defenders find and validate vulnerabilities can also be misused. The trusted-access model shifts part of the safety control from request-level classifiers to customer vetting, program governance and monitored use.
Anthropic’s Mythos product page also positions the model for vetted biology research. Organizations interested in either program should use their account team or Anthropic’s official notification path. A public API ID does not imply that an unapproved account can call the model.
Developer guidance: which Claude model should you use?
| Your situation | Recommended model | Why |
|---|---|---|
| You are an approved Glasswing customer doing defensive cyber or biology research. | Claude Mythos 5 | Use the trusted-access model and the exact integration guidance supplied to your organization. |
| You need Anthropic’s highest broadly released capability for long-running agents, coding or knowledge work. | Claude Fable 5 | Same core model capability with broad-release safeguards; build explicit refusal and fallback handling. |
| You need a broadly available complex-work model at half the Fable/Mythos base token price. | Claude Opus 4.8 | Generally available and positioned for complex agentic coding and enterprise work. Review the confirmed Opus 4.8 availability and API model ID. |
| You need speed and a lower-cost balance for high-volume work. | Claude Sonnet 5 | Anthropic describes it as the current best combination of speed and intelligence. |
Do not use Mythos availability as a reason to postpone an application that does not require its trusted cyber or biology capability. Keep the model ID configurable, test output and tool behavior per model, and make the production default match the access actually granted to the account.
Migrating from Mythos Preview or Opus 4.8
Anthropic provides dedicated migration paths from Claude Mythos Preview to Mythos 5 and from Opus 4.8 to Fable 5. The main integration points to review are:
- replace the model ID with the one explicitly enabled for the account;
- use adaptive thinking and control depth with
effortrather than trying to disable thinking; - do not expect raw chain-of-thought content;
- preserve thinking blocks correctly in multi-turn conversations and when switching models;
- for Fable 5, treat
stop_reason: "refusal"as an expected HTTP 200 response and configure an allowed fallback; - validate 30-day retention against the workload’s data policy before moving sensitive traffic;
- load-test the new tokenizer and price profile with real prompts instead of projecting from Opus token counts.
Keep separate configuration for general_high_capability, complex_production and trusted_research. Map those roles to Fable 5, Opus 4.8 and Mythos 5 only after checking account entitlements. This prevents a limited-access model from becoming an accidental single point of failure.
Check Claude Fable 5 availability and safeguards · Compare the complete current Claude lineup.
FAQ
What is Claude Mythos 5?
Claude Mythos 5 is Anthropic’s limited-availability model for advanced cybersecurity and biology research. It succeeds Claude Mythos Preview and shares Claude Fable 5’s underlying capabilities, specifications and base token pricing.
Is Claude Mythos 5 currently available?
Yes, under limited availability. Anthropic offers Mythos 5 by invitation to approved Project Glasswing customers. There is no self-serve sign-up, and organizations should contact their Anthropic, AWS or Google Cloud account team.
What is the Claude Mythos 5 API model ID?
The current Claude API model ID is claude-mythos-5. Anthropic identifies it as the successor to the earlier trusted-access model claude-mythos-preview.
What is the difference between Claude Mythos 5 and Fable 5?
They share the same underlying model, specifications and pricing. Fable 5 adds safety classifiers for broad availability and can decline or reroute risky requests. Mythos 5 does not include those Fable classifiers and is restricted to vetted trusted-access customers.
How much does Claude Mythos 5 cost?
Anthropic lists base pricing of $10 per million input tokens and $50 per million output tokens, the same as Fable 5. Prompt caching and platform-specific charges should be checked in the current official pricing documentation.
Does Claude Mythos 5 support zero data retention?
No. Anthropic requires 30-day data retention for Mythos 5 and Fable 5 safety monitoring. Teams with zero-data-retention requirements should not route affected workloads to these models.
Should developers use Mythos 5, Fable 5 or Opus 4.8?
Use Mythos 5 only with approved Project Glasswing access for trusted cyber or biology research. Use Fable 5 for the highest broadly released capability with safeguards. Use Opus 4.8 for complex generally available production work when its capability and lower base token price fit the workload.
Update log
- Rebuilt the article around the current Anthropic model catalog: Fable 5 generally available, Mythos 5 limited through Project Glasswing, current model IDs, shared specifications, pricing, retention, API behavior and model-selection guidance.
- Original Claude Mythos explainer published.
Sources and further reading
All current status, access, specification, pricing, retention and API claims were verified only against official Anthropic and Claude Platform sources on July 18, 2026.
- Claude Platform Docs – Models overview.
- Claude Platform Docs – Introducing Claude Fable 5 and Claude Mythos 5.
- Anthropic – Claude Mythos 5 product page.
- Anthropic – Claude Fable 5 product page.
- Anthropic – Project Glasswing.
- Anthropic – Introducing Claude Opus 4.8.
- Claude Platform Docs – Model IDs and versioning.
- Claude Platform Docs – Migration guide.
- Claude Platform Docs – Pricing.
- Claude Platform Docs – Adaptive thinking.
- Claude Platform Docs – Refusals and fallback.
- Claude Platform Docs – API and data retention.

[…] Within hours, multiple GitHub mirrors appeared. Shou’s post on X accumulated over 3 million views. Anthropic pulled the package and deleted older versions, but the internet had already archived everything. This was Anthropic’s second leak in five days — on March 26, a CMS misconfiguration had exposed details about the unreleased Claude Mythos model. […]
[…] those same capabilities collaboratively. The catalyst is a model most of the world has never used: Claude Mythos Preview, which Anthropic describes as a general-purpose frontier system whose coding and reasoning […]
[…] like Claude Mythos signals a new era of AI‑driven cyber attacks and What is Claude Mythos? 7 facts about Anthropic’s leaked AI highlight two big […]
[…] Opus 4.6 (released earlier in Q1 2026) and sits alongside the more broadly capable but restricted Claude Mythos Preview in Anthropic’s current lineup. The model is now generally available — no waitlist, no tier […]