Anthropic Ships Claude Fable 5 — and a Locked-Down Mythos 5 for Cyberdefense Only

Anthropic launched Claude Fable 5 on June 9, calling it a Mythos-class model made safe for general use. The company says Fable 5 exceeds every model it has previously shipped publicly in nearly all benchmarks, with standout performance in software engineering, scientific research, vision tasks, and long-context knowledge work — and the longer and more complex the task, the wider its lead over earlier Claude models.

Alongside Fable 5, Anthropic launched Claude Mythos 5 for a narrow set of recipients. Mythos 5 is the same underlying model, but with the cybersecurity safeguards that apply to Fable 5 removed in certain areas. Access is gated through Project Glasswing, a program run in collaboration with the US government, and Anthropic says Mythos 5 has the strongest offensive and defensive cybersecurity capabilities of any AI model currently available anywhere.

How the Safety Layer Works

Releasing a model with Fable 5's capabilities introduces real risks. Anthropic's approach is to run the model with conservative safeguards that catch queries touching sensitive domains — cybersecurity, biology, and chemistry chief among them. When a query triggers those filters, the response comes from Claude Opus 4.8 instead of Fable 5. Anthropic acknowledges the filters catch some harmless requests too, estimating they trigger in fewer than 5% of sessions on average. The company says it will work to reduce false positives as more capable models make the calibration easier.

The result is a two-tier system: Fable 5 for general use with guardrails, Mythos 5 for organizations — primarily infrastructure providers and cyberdefense teams — who need the full capability set without the restrictions. Anthropic is explicit that Mythos 5 will expand beyond Project Glasswing to a broader trusted access program, but has not given a timeline.

Context Window and Pricing

Both models ship with a 1 million token context window by default and support up to 128,000 output tokens per request — the largest output budget Anthropic has offered. Pricing is $10 per million input tokens and $50 per million output tokens, a significant reduction from the earlier Mythos Preview model. Users on Claude's Pro, Max, Team, and Enterprise subscription plans have access to Fable 5 without extra cost until June 22, after which usage credits apply.

What Mythos 5 Is Already Doing

Anthropic points to early Project Glasswing deployments as evidence of the upside. The Glasswing program — which connected Mythos Preview to US cyberdefense teams — reportedly used the model to help identify and secure vulnerabilities in critically important software infrastructure. In life sciences, Anthropic says early Fable 5 and Mythos 5 deployments have been used to generate novel research hypotheses and accelerate drug candidate development.

The dual-track launch is consistent with Anthropic's stated position: that frontier AI capabilities are both genuinely powerful and genuinely dangerous, and that managing the gap between those two facts requires more than fine-tuning. Releasing Mythos 5 only through a controlled, government-adjacent program while making Fable 5 broadly available is the company's current answer to that tradeoff. According to Anthropic's announcement, covered by The Hacker News and The Guardian among others.