Fable 5 is here. The benchmark flex is real — the moat isn't.

Here's what happened on 9 June: Anthropic opened the gates on its Mythos tier with Fable 5, a model it says beats Opus 4.8 by 10%+ on some benchmarks and works autonomously for longer than anything before it. Good. Genuinely good. This is real engineering: a general-purpose model hitting frontier capability in coding, knowledge work, vision and multi-step research tasks. Anyone who dismisses that is performing cynicism, not doing analysis.

Now the part the launch post soft-pedals. The same Mythos lineage that makes Fable 5 special is the lineage that spooked the government over cyber-vulnerability hunting — so Anthropic fenced the model off in cybersecurity, biology, chemistry and distillation, where it just refuses and hands you Opus 4.8 instead. The most powerful model is the one you're least allowed to point at the hard problems. That's not a product decision — it's a liability management decision dressed up as a product decision.

Free until 22 June, then $10/$50 per million tokens — about double Opus. Anthropic even says paid-plan access only comes back 'when capacity is sufficient', which is how you describe a resource that costs more to run than the previous tier. The pricing tells you something important: Anthropic has chosen to extract margin at the frontier rather than subsidise adoption the way it (and OpenAI) did with earlier models. That's a defensible strategic choice. It's not a moat. A moat is something competitors can't replicate; a price premium is something competitors undercut.

Compare where this stands against the competitive field as of June 2026:

Anthropic's own government-pushback argument — that rivals like OpenAI share the same dangerous capability — is the clearest possible admission that the gap is not what the benchmark chart implies.

You can't simultaneously hold that the dangerous strength is unique enough to fence off and common enough that banning yours is pointless. Pick one. The market already did: capability converges, price competes, and the lead measured in a June benchmark rarely survives to Christmas. The model itself is impressive; the narrative around it is doing more lifting than the model warrants. If you're building agents on top of Anthropic today, the 10% edge is a real and shippable advantage right now. If you're making a long-horizon infrastructure bet, the honest picture is a lab at the frontier, priced at the frontier, with its most capable offering shut down by regulators three days in.

Within 72 hours of launch, the US government issued an export-control directive and Anthropic suspended both Fable 5 and Mythos 5. The fastest route from 'the most capable model we've ever shipped' to 'you can't access it' was three days. That is not a gotcha — regulatory risk at the frontier is real and Anthropic has been more transparent about it than most. But it is a data point about what 'frontier model' means in 2026: a capability that exists inside a political and regulatory container, and the container is smaller than the capability. Whether that resolves into a workable framework or a sustained constraint is the question that actually matters for the next 18 months.