I spent the night of June 9 rebuilding my agent stack — and by morning I'd cleared everything that had been sitting on my task list for weeks. I want to give you the technical picture alongside the personal one.
Two models shipped that night. Claude Fable 5 is the public release — Mythos-class, priced at $10/M input and $50/M output. Mythos 5 is the closed version released under Project Glasswing for cybersecurity defense partners, with loosened operational constraints for specialized threat research.
Fable 5 sits at the top of nearly all the benchmarks we use to evaluate serious coding and reasoning work. Stripe ran their 50M-line Ruby migration through it and compressed the timeline from months to days. In pharma, teams reported up to 10x acceleration on drug development stages. Scientists tested its hypotheses blind — 80% of the time, they chose the model's output.
On safety: dangerous queries trigger a fallback to Opus 4.8, in under 5% of sessions. Anthropic ran over 1,000 hours of adversarial testing and found no universal jailbreak at release.
Pricing context: Fable 5 runs at twice the cost of Opus 4.8, but below half the price of Mythos Preview. Free access through June 22 on Pro, Max, Team, and Enterprise tiers.
The step-change in agentic task completion was immediate — the kind you feel within the first hour. We are living through a genuinely extraordinary moment in how software gets built.
Top comments (0)