From "no plans to release" to public in 9 weeks: the Mythos → Fable 5 timeline
Today's Claude Fable 5 release reads differently if you remember what Anthropic said nine weeks ago. Here's the timeline, with the receipts.
April 7 · Mythos Preview: "we have no plan to release this"
Anthropic disclosed Claude Mythos Preview, a general-purpose model that turned out to be strikingly good at offensive security. The numbers that mattered:
- Where Opus 4.6 had a near-0% success rate at autonomous exploit development, Mythos Preview produced 181 working Firefox exploits (Opus 4.6: 2).
- It found zero-days in OpenBSD (including a 27-year-old TCP SACK vulnerability), plus FreeBSD, Linux, every major browser, and a 16-year-old bug in FFmpeg's H.264 codec.
- It autonomously wrote a JIT heap spray chaining four vulnerabilities to escape a browser sandbox.
Anthropic's response was not to ship it. Instead they launched Project Glasswing: limited access for critical industry partners and open-source developers to find and fix vulnerabilities before any broader release. Over 99% of what Mythos found remains unpatched, so disclosure is happening through SHA-3 commitment hashes with full details following patches.
June 2 · Glasswing scales up
TechCrunch reported Anthropic scaling Mythos access to critical-infrastructure operators in 15+ countries. In hindsight, this was the dress rehearsal: validating that the patch pipeline and access controls worked at scale.
June 9 · Fable 5: the same model, gated
Today the model went public under a new name. The key design decisions:
Same weights, different gates. Fable 5 is Mythos 5. The public version adds classifier-based safeguards on three fronts: offensive cyber, dual-use biology/chemistry, and distillation attempts.
Fallback, not refusal. Tripping a safeguard routes your query to Claude Opus 4.8 instead of returning a refusal. Under 5% of sessions see it.
Tiered trust. Glasswing partners get Mythos 5 with cyber safeguards lifted. A biology-researcher tier is forthcoming. Everyone else gets Fable 5, which is still, by Anthropic's account, state-of-the-art on nearly every benchmark they tested.
The price came down. $10/$50 per million tokens, less than half of Mythos Preview pricing.
The pattern worth noticing
The Mythos rollout is the first time a frontier lab has openly run the sequence capability discovery → restricted deployment → infrastructure hardening → gated public release on that capability. Whether you think nine weeks is fast or slow, the sequence itself is now precedent. Expect it again.
For working engineers the practical takeaway is simpler: the most capable coding model on the market is now an API call away, and the design assumption baked into it is long-horizon autonomous work. The bottleneck is shifting from "can the model do it" to "can you specify, checkpoint, and verify it."
That last part is what Urbano DX does for clients: scoping AI capabilities into shippable software as paid proof-of-concepts, production APIs, and AI workflows. If you're trying to figure out what Fable 5-class models change for your product, start a conversation.