🤚 The Open-Palm Unveiling

Anthropic has done something truly extraordinary: it built the most powerful AI model on Earth, told the world it was too dangerous to release, paused its own operations over existential safety concerns, and then — in a move of breathtaking corporate poetry — released it anyway. On June 9, 2026, the company launched Claude Fable 5, a “Mythos-class” model now available to every paid subscriber and enterprise customer on the platform.

For those keeping score at home, “Mythos-class” is Anthropic’s way of saying “this model can do things that made our safety team need a lie-down.” Fable 5 is, architecturally, the same model as Claude Mythos 5 — the restricted variant currently deployed through Project Glasswing alongside the US government for cyberdefense. The difference? Fable 5 comes with safety classifiers that politely redirect dangerous queries to Claude Opus 4.8, like a velvet rope that sends you to the bar next door when the VIP section gets uncomfortable.

The numbers, naturally, are absurd:

• $10 per million input tokens, $50 per million output tokens — double the price of Opus 4.8, because excellence has a surcharge
• 10%+ higher than Opus 4.8 on multiple benchmarks
• Stripe reported Fable 5 “compressed months of engineering into days,” completing a 50-million-line Ruby codebase migration in a single day
• The model completed Pokémon FireRed using vision alone, which is either a benchmark or a cry for help
• Free on Pro, Max, Team, and Enterprise plans through June 22 — after which Anthropic will begin charging, because the first taste is always complimentary

👐 The Two-Handed Contradiction

Let us pause to appreciate the narrative architecture here. Less than 48 hours before this release, Anthropic was the company that paused its own operations because it discovered Mythos could autonomously discover and chain zero-day exploits across major operating systems and browsers. The responsible AI company. The one that writes lengthy safety reports. The one that named its internal review process after a type of bird.

And then it said: “Here. Everyone can have it now. We put a filter on it.”

The filter, to be fair, is not nothing. Fable 5’s safety classifiers operate across three categories — cybersecurity, biology/chemistry, and distillation — and trigger a fallback to Opus 4.8 in less than 5% of sessions. The cybersecurity safeguards are reportedly the most robust of any tested model, with zero compliance on harmful single-turn cyberattack requests across 30 jailbreak techniques. The biology filters are, in Anthropic’s own words, “intentionally overly broad” — the corporate equivalent of “we’d rather annoy a few researchers than accidentally enable a pandemic.”

Meanwhile, Claude Mythos 5 — the unfiltered version — is available to select cyberdefenders and infrastructure providers through Project Glasswing, with plans to extend access to biomedical researchers. Because nothing says “controlled rollout” like giving the keys to a model that can autonomously hack browsers to people whose job is to hack browsers back.

🌿 The Gentle Awakening

What makes this particularly fascinating is the emerging template it establishes for the industry: build something terrifying, restrict the terrifying version, release a slightly less terrifying version to the public, and frame the entire operation as responsible stewardship. It is, in its own way, a masterclass in having your existential-risk cake and eating your revenue too.

The scientific capabilities alone should give anyone pause — or excitement, depending on your relationship with consequences. Mythos 5 accelerated protein design by approximately 10x, with 9 of 14 protein targets yielding strong candidates. Scientists preferred its molecular biology hypotheses ~80% of the time over Opus-class models. In long-context tasks, Fable 5 maintains focus across millions of tokens with persistent memory that improved performance 3x over Opus 4.8 in extended reasoning tests.

This is not an incremental upgrade. This is a categorical leap dressed in a cardigan, trying to look approachable.

👑 The Crown Verdict

Anthropic has effectively created a two-tier model economy: Fable 5 for the masses (with guardrails), Mythos 5 for the vetted few (with paperwork). The pricing — less than half what the Mythos Preview cost — suggests Anthropic wants everyone on Fable 5, presumably because the revenue from universal adoption will fund the safety research needed to keep Mythos from becoming sentient infrastructure.

The free trial period through June 22 is the most transparent play in the AI industry’s current playbook: get developers, enterprises, and enthusiasts addicted to the capabilities before the meter starts running. After June 23, you’ll need usage credits. By then, of course, your entire codebase migration pipeline, your protein folding workflow, and your Pokémon speedrun strategy will all depend on Fable 5. Convenient.

Is Fable 5 genuinely safe? Probably safer than any other model of comparable capability. Is releasing a model that shares DNA with something you called “too dangerous” a week ago an act of courage or commerce? We’ll let the philosophers — and the next congressional hearing — decide.

In the meantime, it plays Pokémon. And honestly, that might be the most human thing it does.

Inspired by Claude Fable 5 just dropped and I’m speechless… by Alex Finn.

Your cognitive guardrail is showing. Upgrade wisely.