Anthropic this week unveiled Claude Fable 5, positioning it as the company’s most capable Mythos-class model yet and rolling out new safety controls aimed at reducing misuse. The model is available now across Anthropic’s Claude products and via the Claude API under the identifier claude-fable-5. Anthropic also introduced Claude Mythos 5 for a small, approved set of cyberdefenders and critical infrastructure providers.
What’s new and notable
- Capability: Anthropic says Fable 5 handles longer, more complex and more autonomous tasks than prior Claude models—able to stay focused across “millions of tokens” and to operate for extended, multi-step workflows. The company highlights gains in software engineering, knowledge work, vision, finance, memory and scientific research, with the biggest lead on longer, complex tasks.
- Real-world example: Stripe reported that Fable 5 completed a large Ruby migration in a single day—a job Anthropic says would have taken a team over two months by hand.
- Access: Developers can call the model via the claude-fable-5 API identifier; Fable 5 is included in Pro, Max, Team and seat-based enterprise plans through June 22. After June 23, usage will require credits unless capacity allows an extension.
New safeguards and fallbacks
Anthropic built extra safety controls given Fable 5’s expanded capabilities. Some requests—primarily those touching sensitive domains—will automatically fall back to Claude Opus 4.8. The fallback covers selected cybersecurity, biology, chemistry and distillation-related queries and is designed to block potential misuse; Anthropic says these safeguards trigger in under 5% of sessions on average, meaning more than 95% of sessions run without fallback.
The company admits the filters are conservatively tuned and may capture some harmless queries; it plans to adjust them post-launch to reduce false positives. New classifiers are in place to catch misuse and jailbreak attempts and to prevent Fable 5 from responding directly to flagged requests. The cybersecurity controls specifically target exploitation and offensive cyber tasks, while biology/chemistry safeguards reflect dual-use risk concerns.
Claude Mythos 5 and trusted access
Anthropic also launched Claude Mythos 5 for a narrower, vetted audience. Mythos 5 uses the same underlying model as Fable 5 but is being distributed through Project Glasswing in cooperation with the U.S. government to approved cyberdefenders and critical software infrastructure providers. For those approved cybersecurity users, some safeguards are relaxed; Anthropic also plans a trusted access program for selected biology researchers and companies.
Data handling and pricing
Anthropic said business customer traffic on Mythos-class models will face 30-day retention and that this data will not be used to train future Claude models. Pricing is set at $10 per million input tokens and $50 per million output tokens.
Why this matters to crypto and web3
For crypto firms—exchanges, custodians, smart-contract audit teams, on-chain analytics providers and infrastructure operators—Fable 5’s longer-context reasoning and automation could accelerate code migrations, contract analysis, incident response and large-scale data synthesis. The Mythos program’s focus on vetted cyberdefenders and the 30-day retention policy will also be particularly relevant to organizations balancing high-assurance security needs with data privacy and compliance concerns.
Bottom line: Anthropic is betting on bigger scale and longer-lived workflows with Fable 5, while layering conservative safeguards and a vetted access path (Mythos 5) for sensitive users. The model’s capabilities and pricing make it worth watching for crypto infra and security teams evaluating how to incorporate advanced LLMs into audits, migrations and defenses.
Read more AI-generated news on: undefined/news
