What A Time To Be ALIVE!

Anthropic shipped Claude Fable 5 today, the first publicly available model in its new Mythos-class tier — the rung that sits above Opus. Remember the "Mythos-class models in the coming weeks" line buried at the bottom of the Opus 4.8 announcement we flagged on May 28? This is that. The timeline held.

It's available immediately through the Claude API and usage-based Enterprise plans, and it is the first model in the Claude 5 family. The story most outlets are running is the obvious one: "Anthropic's most powerful public model yet, crushes the benchmarks." That's true, and it's fine. But it misses what's actually interesting.

Fable 5 is the first major frontier release where the headline feature is what the model refuses to do. This is the same Mythos brain that Anthropic deemed too dangerous to release back in April — the one that set off a worldwide cybersecurity scramble because it was so good at finding and exploiting software bugs. Anthropic's answer wasn't to make it less capable. It was to build a public version that hard-blocks the scary stuff and routes it to the older, tamer Opus 4.8. The capability race and the safety story are now the same product. That's the shift.

Here are the verified facts and which ones matter most for your work.

It's a real generational jump, not a point release. Anthropic says Fable 5 beats every model it has ever made generally available, and claims state-of-the-art on nearly all tested benchmarks. The gap widens the longer and more complex the task gets. This is the first Claude 5 model, not a tweak to the 4 line.

The benchmarks are not close. SWE-bench Pro hit 80.3% (Opus 4.8 lands at 69.2%, GPT-5.5 at 58.6%, Gemini 3.1 Pro at 54.2%). On Cognition's FrontierCode — brutal production-grade coding — Fable scored 29.3% versus Opus 4.8's 13.4% and GPT-5.5's 5.7%. That FrontierCode line is the eye-popper: it more than doubles Opus and laps everyone else.

The guardrails are the actual product. Ask Fable 5 anything in cybersecurity, biology, chemistry, or model distillation, and it won't answer — it defers to Claude Opus 4.8. The bio and cyber blocks are the unspoken headline: these are the exact capabilities that kept Mythos locked up. Anthropic says its red team spent 1,000+ hours hunting for a jailbreak around these blocks and couldn't find a single working one.

It costs double. Fable 5 runs about 2x Claude Opus 4.8 — roughly $10 per million input tokens. For most upgrade cycles a price hike is hard to justify; here Anthropic is betting the capability jump speaks for itself. If you're running volume, this is the number to model out before you switch anything.

Mythos 5 shipped too — but not for you. Alongside Fable, Anthropic released Claude Mythos 5, the uncuffed sibling, to trusted partners only. It shows strong results in drug design and runs largely autonomously in genomics research. The gap between what Fable won't do and what Mythos can is the real story of this generation.

The framing. Anthropic's product-research lead Dianne Penn called it a "race to the top" — ship the powerful thing, wrap it so it delivers "asymmetrically more benefit than harm."

For an audience that has deployed AI in production, this is the change that reframes the buying decision.

Every lab is selling raw capability. Anthropic just shipped a flagship whose pitch is partly what it declines to touch — and made that the reason it's safe enough to release at all. Two months ago this exact capability level was considered too dangerous for the public. Today it's a paid product. The thing that changed wasn't the model's power; it was Anthropic's confidence in the containment.

For enterprise buyers, that's the whole game. The customers writing seven- and eight-figure checks — the banks, the consultancies, the critical-infrastructure operators Anthropic spent the spring onboarding to Mythos — don't lose sleep over a 2% benchmark delta. They lose sleep over deploying a model that can design a bioweapon or auto-discover zero-days. Fable 5 is the first frontier model engineered so the answer to "what happens if this gets misused" is a built-in wall instead of a policy promise. That's why Anthropic can charge double and still expect the enterprise pipeline to grow.

Read this part carefully, because the free window is short.

Through June 22, Fable 5 is bundled at no extra cost into Pro, Max, Team, and seat-based Enterprise plans. Go play. On June 23, Anthropic pulls it from those plans after that you'll burn usage credits to keep using it, with a plan to fold it back into standard subscriptions "as soon as possible." Translation: you've got about two weeks of free joyriding before the meter starts.

It's available right now through the Claude API and consumption-based Enterprise plans if you don't want to wait on the subscription tiers.

One related kicker: this launch lands days after Anthropic publicly urged the major labs to agree on a coordinated "brake pedal" for frontier AI, warning that systems may be approaching recursive self-improvement. The same company telling the industry to slow down just shipped its most powerful public model and priced it at a premium. That's not a contradiction, it's the entire Anthropic personality compiled and shipped.

For working developers, Fable 5 is a clear upgrade for hard, long-horizon coding the FrontierCode and SWE-bench Pro gaps are large enough that you'll feel them on real refactors and migrations. Just know it'll bounce you to Opus 4.8 the moment your task brushes against security or bio work, which matters if that's your domain.

For teams modeling cost, the 2x price is the catch. Run a real comparison this week: the capability jump may pay for itself on your hardest tasks, but it almost certainly isn't worth double on routine work where Opus 4.8 already clears the bar. Use the free window through June 22 to benchmark it against your actual workloads before the credits kick in.

For anyone watching the frontier race, today is the cleanest signal yet of Anthropic's bet: the next axis of competition isn't just capability, it's deployable capability, power that's been made safe enough to sell at scale. Every other lab is racing to release more. Anthropic just proved its edge is being the one that can release the dangerous thing at all. That's the bet that's been repricing this company.

Claude Fable 5 shipped today as the first publicly available Mythos-class model and the first in the Claude 5 family, hitting exactly the "coming weeks" timeline Anthropic teased in the Opus 4.8 announcement. It's state-of-the-art on nearly every benchmark (SWE-bench Pro 80.3%, FrontierCode 29.3%, more than double Opus 4.8), priced at roughly 2x Opus at ~$10 per million input tokens, and hard-blocks cybersecurity, biology, chemistry, and distillation requests by deferring them to Opus 4.8. The red team logged 1,000+ hours without a working jailbreak. Mythos 5, the uncuffed version, went to partners only. It's free in Pro/Max/Team/seat Enterprise through June 22, then moves to usage credits on June 23. The version number says Fable 5. The actual story is that Anthropic figured out how to sell the model it was too scared to release, and that's the moat.

Stay building. 🤖