How to Get Claude Mythos 5 Access (June 2026)

Claude Mythos 5 is Anthropic’s frontier model without the cybersecurity classifiers and content safeguards that ship with its sibling Claude Fable 5. It’s not publicly available. Access is restricted to approved partners — mostly through Project Glasswing for cybersecurity and a separate program for biology/chemistry research. This page explains exactly who can access it, how to apply, and what to use instead if you can’t.

Last verified: June 15, 2026.

TL;DR

• Mythos 5 is NOT generally available. No public sign-up.
• Path 1: Project Glasswing (cybersecurity partners). Apply via your Anthropic, AWS, or Google Cloud account team.
• Path 2: Trusted biology / chemistry research program. Apply via [email protected] with a written protocol.
• For everyone else: Use Claude Fable 5 — same core capabilities, with safeguards. Available on Claude API, AWS Bedrock, Google Vertex AI, Microsoft Foundry from June 9, 2026.

What Mythos 5 actually is

Claude Mythos 5 and Claude Fable 5 were announced together on June 9, 2026, as the first models in Anthropic’s new Mythos-class tier — a capability class that sits above the Opus class. Anthropic’s framing: Mythos-class models have crossed a threshold where additional safeguards became necessary for general release.

Same model, two configurations:

	Claude Fable 5	Claude Mythos 5
Public availability	Yes (June 9, 2026)	No — restricted partners only
Core weights	Same	Same
Cybersecurity classifier	Yes	No
Bio/chem additional restrictions	Yes	Reduced
Standard content safeguards	Full	Reduced
API model ID	claude-fable-5	claude-mythos-5
Price	$10/$50 per Mtok	$10/$50 per Mtok
Cache read	$1/Mtok	$1/Mtok
Where to access	Claude API, AWS Bedrock, Google Vertex AI, Microsoft Foundry	Restricted-access channels only

Anthropic has been explicit that the capability ceiling is identical. The difference is the safeguard layer that screens model outputs.

Path 1: Project Glasswing (cybersecurity)

Project Glasswing is Anthropic’s program for trusted cybersecurity partners. The named focus is enabling legitimate offensive-security research — penetration testing, red-team exercises, vulnerability research — that the public Fable 5 safeguards would block as false positives.

Who has been accepted (publicly disclosed or reported):

• Major cybersecurity firms with established red-team practices
• Select cloud providers and security vendors
• Government cybersecurity agencies in allied countries (subject to additional review)
• Large financial institutions with internal red-team capability

How to apply:

1. Have an existing Anthropic Enterprise account, or AWS / Google Cloud relationship that includes Claude access.
2. Contact your account team and request consideration for Project Glasswing.
3. Provide a written description of your security research use case, organizational red-team program, and proposed safeguards.
4. Sign Anthropic’s supplemental Mythos-access agreement covering responsible use, incident reporting, and ongoing audit cooperation.
5. Wait 4–12 weeks for review.

What disqualifies you: No documented red-team / security research program. Individual researchers without organizational backing. Use cases that look like general-purpose access dressed up as security research.

Path 2: Biology and chemistry research

Anthropic has a separate trusted-access program for biology and chemistry researchers. The bar is even higher because Mythos 5’s reduced bio safeguards are the most carefully gated part of the access policy.

Typical accepted profile:

• Established academic research labs at universities with institutional biosafety review
• Pharmaceutical companies with internal compliance programs
• Public-interest biology research organizations

How to apply:

1. Email [email protected] with a written research protocol, IRB or equivalent biosafety review documentation, and institutional affiliation.
2. Provide a clear scientific justification for why reduced safeguards are necessary (i.e., why Fable 5’s defaults block legitimate research).
3. Anthropic reviews against published criteria and may request additional documentation.
4. If accepted, sign a supplemental research-use agreement.

This path is reportedly stricter than Project Glasswing. Expect rejections for individual researchers without institutional review processes.

Path 3: What Anthropic has hinted may come later

In its June 9, 2026 announcement, Anthropic signaled that additional restricted programs may follow in areas including:

• AI safety research labs
• Frontier AI evaluation organizations
• Government scientific research institutions

These programs do not exist publicly as of June 15, 2026. Anthropic’s framing is that it will open additional access channels as it gains operational experience with Mythos 5 and as the underlying safety classifiers mature.

What to use if you can’t get access

For 95%+ of legitimate use cases, Claude Fable 5 is the right answer:

• Same frontier capabilities (80.3% SWE-Bench Pro, leading the Artificial Analysis Intelligence Index)
• Same price ($10/$50 per Mtok)
• Same 1M token context window
• Generally available on Claude API, AWS Bedrock, Google Vertex AI, Microsoft Foundry from day one

The published refusal-rate delta between Fable 5 and Mythos 5 on developer workloads is small — Fable 5 refuses roughly 3–5% more queries, and most of those refusals are appropriate (genuine policy violations) rather than false positives.

If you hit a specific Fable 5 refusal on a legitimate workflow:

1. Document the prompt, response, and your use-case justification.
2. Email your Anthropic account contact (or [email protected]).
3. Anthropic has a per-use-case whitelisting process that can grant exceptions without full Mythos access. Lead time: 1–4 weeks.

This is the right path for most teams. Full Mythos access is overkill if you only have a handful of false-positive refusals per month.

Common misconceptions

• “Mythos 5 is more capable than Fable 5.” False. Same core weights, same capabilities. Mythos has fewer refusals; that’s the only meaningful difference for end-user experience.
• “I can sign up for Mythos with a credit card.” False. There is no public sign-up. Access is granted through direct partnership review.
• “AWS or Google Cloud sells Mythos.” Partial. Mythos can be accessed via AWS Bedrock or Google Vertex AI, but only for accounts that have been approved through Project Glasswing or the research program; the public Bedrock/Vertex catalogs only expose Fable 5.
• “Mythos 5 doesn’t have any safety filters.” False. Mythos 5 retains Anthropic’s baseline constitutional AI training and the Acceptable Use Policy. What’s removed is the additional cybersecurity classifier and certain expanded restrictions added to Fable 5.

Pricing comparison

Cost item	Mythos 5	Fable 5	Opus 4.8
Input	$10/Mtok	$10/Mtok	$5/Mtok
Output	$50/Mtok	$50/Mtok	$25/Mtok
Cache read	$1/Mtok	$1/Mtok	$0.50/Mtok
Batch discount	50%	50%	50%
Plan inclusion	None (credits-only)	Pro/Max/Team until June 22, then credits	Pro/Max/Team

Mythos 5 doesn’t get plan-included access — it’s pure metered credits or consumption-based Enterprise — because the partner program model bypasses subscription plans entirely.

Timeline expectations

• Project Glasswing application: 4–12 weeks from initial request to API access.
• Research program application: 6–16 weeks; expect more documentation back-and-forth.
• Per-case Fable 5 whitelisting: 1–4 weeks (the right answer for most narrow use cases).
• Wait-and-see for broader access: No public timeline. Anthropic has historically expanded restricted programs incrementally; expect months to quarters, not days.

Related reading

• Claude Fable 5 vs Mythos 5 vs GPT-5.5: Public Release (June 2026)
• Claude Fable 5 Cybersecurity Restrictions Explained
• Claude Mythos Preview vs Claude Opus 4.6 (April 2026)
• Should You Wait for Claude Mythos Public Release? (June 2026)

---

Anthropic’s restricted-access programs change over time. Confirm current criteria with your Anthropic account team before basing a procurement decision on this guide.