Anthropic just released its most capable model ever.
Anthropic just released its most capable model ever. The pricing tells you more than the benchmarks.
Today: Claude Fable 5 — the first Mythos-class model made generally available. Same brain as Mythos 5, the restricted model that found vulnerabilities across every major OS during Project Glasswing. Now with guardrails fitted for public roads.
Three signals worth your attention:
▸ Capability routing as governance. Ask Fable 5 something high-risk in cyber or bio, and it doesn’t refuse — it silently falls back to Opus 4.8. One product, two models, policy enforced at the routing layer. This is exactly the architecture pattern enterprises will need internally: capability tiers, not capability bans.
▸ Tiered trust as distribution model. Mythos 5 — same model, safeguards lifted — goes only to vetted cyberdefenders and infrastructure providers, expanding via a trusted-access programme. Access to frontier capability is becoming a credential, not a credit card. Partner ecosystems should take note.
▸ Premium pricing as a statement. 10 USD in / 50 USD out per million tokens — roughly 2× Opus 4.8, and Anthropic expects demand to outstrip capacity anyway. When the most expensive model on the market is also the most demanded, the unit of value has shifted: from tokens consumed → outcomes delivered. Stripe reports months of engineering compressed into days. That’s not a tooling upgrade. That’s a billing-model question.
The benchmark numbers are impressive (80,3 % on SWE-Bench Pro vs 58,6 % for GPT-5.5). But the real headline is autonomy: longer unsupervised runs than any previous Claude, holding focus across millions of tokens.
The models are ready to work for days, not seconds.
Are your operating model and your pricing model?
Originally posted on LinkedIn on 2026-06-09
Enjoyed this article?
Let's continue the conversation on LinkedIn
I share weekly insights on AI strategy, enterprise transformation, and lessons from nearly thirty years in the field.
Follow Magnus on LinkedIn