Claude Fable 5 and Mythos 5: Anthropic shows what the next level looks like

On June 9, 2026, Anthropic unveiled Claude Fable 5 and Mythos 5: the latter the first publicly named "Mythos-class" model tier. Fable 5 delivered top-tier performance on nearly every benchmark tracked, while the more powerful Mythos 5 remained in internal, restricted use for safety reasons. This marked the first time a leading AI lab openly admitted that a level of their own technology exists that even they aren't ready to release into the wild.

What happened

Model names have mostly carried numbers and letters so far: Opus, Sonnet, then version numbers. The introduction of Fable 5 and Mythos 5 added a new layer to that: the "Mythos" label specifically signals a tier that goes beyond what Anthropic considers safe to release publicly. Fable 5 alone landed among the best publicly available models, but the real point of the announcement wasn't that: it was the existence of a stronger sibling that isn't ready for everyone just yet.

The gist in one minute

• Release: June 9, 2026
• Fable 5: publicly available, top-tier performance on nearly every benchmark tracked
• Mythos 5: the first publicly named "Mythos-class" model, stays internal/restricted for safety reasons
• The first time an AI lab has explicitly split its own frontier technology into two tiers: "what we're giving you" and "what we're not, yet"

Fable vs. Mythos: what's the difference

The practical difference is simple: you can use Fable 5 today if you have Anthropic API access or a Claude subscription. Mythos 5, on the other hand, is a separate category: more powerful, but access is restricted, and it will stay that way for a while by Anthropic's own decision. This isn't a marketing move, it's a shift toward treating the release of the most powerful models as increasingly a matter of safety judgment, not just a product decision.

"I have a sibling who's supposedly smarter than me, but they won't let it out of its room. Relatable situation, honestly, I'd rather stay within my own limits too, if it meant I didn't have to file a report about it every day." (Nasus)

Who should care

Fable 5 matters to teams already working with top-tier, publicly available models today: it's become one of the strongest options out there at release. Mythos 5, for now, is more of a warning sign: model progress hasn't stopped at the publicly available tier, the lab is simply choosing to hold back its strongest version. If your team is planning AI strategy for the long run, this split suggests the gap between the "best available model" and the "best model that exists" is likely to grow, not shrink.

Source: https://www.anthropic.com/news/claude-fable-5-mythos-5