Anthropic released Claude Fable 5 on June 9, 2026, and it is not another point release. It is the first publicly available model in the Claude 5 family and the first to carry what the company calls a Mythos-class capability tier. That tier sits above Opus, which had been the top of the line since 2024. If you have been waiting for the next real jump rather than another decimal, this is it.
A new tier, not a version bump
For two years the Claude lineup followed a three-step ladder: Haiku for speed, Sonnet for balance, Opus for the hard problems. Fable 5 starts a fourth step. Anthropic's announcement describes it as a Mythos-class model, the bracket it created for the architecture behind Claude Mythos 5, a system the company has kept under restricted access because of its cybersecurity capabilities.
The naming choice tells you how Anthropic thinks about it. Fable 5 is not called Opus 5. It shares its underlying model with Mythos 5, and the two differ only in the safeguards wrapped around them. Same brain, different rules.
How the safeguards work
Fable 5 routes every conversation through classifiers, separate AI systems that watch for potential misuse. When a query touches offensive cybersecurity, certain biology and chemistry territory, or attempts to distill the model, Fable 5 does not answer it. Claude Opus 4.8 answers instead. Anthropic says more than 95% of Fable sessions involve no fallback at all, so for most work you will never notice the seam.
Two details matter if you handle sensitive material. First, the classifiers are tuned to be cautious, so an occasional benign request will trip them. Second, all traffic on Mythos-class models carries a 30-day retention policy. We unpack both, and what the unrestricted sibling can do, in our look at the Fable 5 guardrails.
What it is actually good at
The headline capability is long-horizon work: tasks that run for hours and span many steps without a human nudging the model back on course. The early evidence is concrete:
- Coding at scale. Fable 5 scored 80.3% on SWE-Bench Pro, against 69.2% for Opus 4.8. Stripe reports it compressed a codebase-wide migration that would have taken a team over two months into a single day.
- Sustained analysis. It is the first model to clear 90% on Hex's analytical benchmark, built from long, multi-stage data tasks, and it posted the highest score of any model on Hebbia's finance benchmark.
- Memory that compounds. Given persistent memory in the game Slay the Spire, Fable 5 improved three times more than Opus 4.8 did under the same setup.
- Science workflows. Anthropic reports it accelerated parts of drug design work by around ten times.
The full benchmark picture, including where the gains are thinner, is in our Fable 5 versus Opus 4.8 versus GPT-5.5 comparison.
What it costs and where to get it
Fable 5 is priced at $10 per million input tokens and $50 per million output tokens, exactly double Opus 4.8. It is live on the API today as claude-fable-5, and Anthropic is including it on Pro, Max, Team, and seat-based Enterprise plans through June 22. After that, subscription users need usage credits. The practical decision of when Fable 5 earns its premium is covered in our pricing and access guide.
Why this launch matters
Fable 5 arrives eight days after Anthropic confidentially filed for an IPO, and two months after the company chose not to release this architecture publicly, routing it instead into Project Glasswing, a defensive security program whose partners include Microsoft, Google, and NATO-affiliated agencies. The model you can use today found a 27-year-old OpenBSD flaw in its unrestricted form. That backstory, from leak to launch, is a story worth reading in full.
"We're developing a general purpose model with meaningful advances in reasoning, coding, and cybersecurity. Given the strength of its capabilities, we're being deliberate about how we release it." That was Anthropic in March, confirming the model's existence after a leak. June 9 is what deliberate looked like.
The short version: a new capability bracket is now public, the safety story is unusually transparent, and the price doubles rather than multiplies. For developers and teams doing long, hard work with AI, today changed the menu.