Anthropic released Claude Fable 5 and Claude Mythos 5 on June 9 — one model class, split into two tiers by access rather than by size. Mythos 5, the unrestricted-capability tier, is limited to Project Glasswing partners and select biology researchers. Fable 5 is the version everyone else gets: in Anthropic's words, "a Mythos-class model that we've made safe for general use." It's generally available at $10 per million input tokens and $50 per million output.

The launch claims are vendor claims, so hold them at arm's length. Anthropic says "Fable 5 scores highest among frontier models, even at medium effort," and that "Fable 5 has the highest score of any model on Hebbia's Finance Benchmark." For the restricted tier, Anthropic says "Mythos 5 executed tasks normally completed by a scientist with a proficiency that matches or beats skilled human operators in drug design," and that "Mythos 5 is our first model to consistently produce novel, compelling scientific hypotheses." Day-one numbers, independently unverified — the same rule we apply to every launch page.

One claim is different in kind, and it's the one worth your attention. Anthropic says "Fable 5 performed a codebase-wide migration in a day that would otherwise have taken a whole team over two months by hand."

A day versus two months

That sentence isn't a benchmark score. It's a statement about work — the clearest dollars-and-weeks framing any vendor has given for what long-horizon autonomous AI actually means. Put rough numbers on it: a four-person engineering team for two months is on the order of $150,000 in loaded payroll. A day of model time, even at $10 and $50 per million tokens with heavy usage, lands in the hundreds of dollars. Discount the claim by three quarters and the economics of project-sized work — migrations, system rewrites, the cleanup nobody funds — still moved.

Two cautions keep that honest. First, it's one example from the vendor's own launch page, not a repeatable price list; your codebase and your two months are not theirs. Second, a day of autonomous work still needs a person who can specify the job precisely and verify the result — Anthropic also says "Fable 5 can autonomously complete multi-agent workflows with fewer turns and 25-30% faster than prior models," which is a speed claim, not a supervision waiver.

The practical posture is the one we take anyway: match the tool to the task, not the launch page to the roadmap. Most of what a 2–50 person business should automate first doesn't need a frontier model at all — that's the point of Solution Levels. What changed is the ceiling on what's automatable at any level.

Three days to export controls

Then the part with no precedent in a model launch. On June 12 — three days after release — the US government applied export controls to both models. On June 30, Anthropic published "Redeploying Fable 5": Fable 5 returned globally on July 1, and the export controls on Mythos 5 were lifted the same day, with enhanced safeguards around cybersecurity and chemistry.

Read plainly, the timeline says two things. Top-end AI now has a regulatory dimension — access to the strongest models can change on a government's schedule, not just a vendor's. And the disruption was bounded: nineteen days from controls to resolution. That's not a reason to avoid frontier models. It's a new line item on the risk list, the same way a sole-source supplier is.

What to do with this

  • Build on the generally available tier. Fable 5, and the affordable tier of the same generation below it (Claude Sonnet 5), is where pricing and terms are published and stable. Restricted-access tiers aren't a planning input for a small business — and even the general tier lost nineteen days this June.
  • Treat model availability like supplier risk. If a workflow depends on one model, name the model you'd switch to and test the switch before you need it. Three days from launch to export controls is now a real data point, not a hypothetical.
  • Re-run the math on parked, project-sized work. The two-months-to-a-day claim, discounted as hard as you like, changes what belongs on the "someday" list. Run candidates through the same three-question filter before scoping anything.
  • File Mythos 5 as a signal, not a tool. Unless you're a Project Glasswing partner or a biology researcher, you can't buy it. Its job in your planning is to show where the capability ceiling is headed — and that governments are now watching that ceiling too.

If you're deciding what frontier-tier capability is worth to your business — and what stable ground to build on while the rules settle — that's a 30-minute conversation.

Sources: Claude Fable 5 and Claude Mythos 5 — Anthropic, June 9, 2026 · Redeploying Fable 5 — Anthropic, June 30, 2026