Overview
Claude Mythos 5 is a **research-restricted version** released by Anthropic on June 9, 2026, simultaneously with Fable 5. It shares the exact same underlying model as Fable 5, with the only difference being that **some safety guardrails are removed**—primarily targeting professional scenarios such as biomedical research and cyber defense.
It is not a publicly available subscription product; access is through **Project Glasswing** (a collaboration between Anthropic and the U.S. government launched in April 2026, now expanded to approximately 150 organizations across 15 countries) and the gradually launched "trusted access program." Existing partners of Mythos Preview can directly upgrade to Mythos 5.
Key Features
- 10x acceleration in drug design: Some parts of the protein design pipeline are accelerated by approximately 10x; 9 out of 14 targets produced strong drug candidates (including AAV candidates, in collaboration with Dyno Therapeutics)
- SOTA in molecular biology hypothesis generation: In blind tests, Anthropic scientists preferred Mythos-generated hypotheses about 80% of the time; one E. coli protein hypothesis has been validated by an independent lab
- Autonomous genomics research: Autonomously worked for over a week across 138 animal species to train ML models, outperforming a recent model published in Science with only 1/100 of its parameters
- Open cybersecurity research: Project Glasswing partners can use Mythos 5 for defensive research in evaluations such as Firefox / OSS-Fuzz / CyberGym / CyScenarioBench
- Removal of some guardrails: Biochemical and cybersecurity guardrails are removed for trusted researchers, enabling highly sensitive research within compliance frameworks
- 30-day data retention + no training: Mythos-level traffic is uniformly retained for 30 days solely for security auditing, and explicitly not used for training new models
Use Cases
- Cyber defense research (vulnerability discovery, defense architecture design)
- Biomedical research (protein design, molecular hypothesis generation, genomics)
- Critical infrastructure protection for trusted enterprises
- Government/research institutions requiring Mythos-level capabilities with compliance qualifications
Pros
- The strongest publicly available Claude model (including capabilities not unlocked in Fable 5)
- Verifiable real-world research outputs in fields like biomedicine/genomics
- Transparent data retention policy (30 days + not used for training)
- Leverages the established 150-organization partner network of Project Glasswing
- Same underlying model as Fable 5, equivalent capabilities
Pricing
Mythos 5 does not follow standard pricing; access must be applied for through Project Glasswing or the subsequent trusted access program. Specific business terms are negotiated on a case-by-case basis; existing Mythos Preview partners can upgrade directly.
Summary
Mythos 5 is not a product for ordinary users—it is Anthropic's way of selectively granting the 'strongest Claude' to trusted groups. If you are in cyber defense, critical infrastructure provision, or compliant biomedical research, pay attention to the rollout of Project Glasswing and the trusted access program. For regular developers and enterprise users, Fable 5 suffices (same underlying model, just with three additional guardrails). Anthropic's strategy is 'capability grading, safety layering'—Mythos 5 is the cutting-edge sample of this strategy.