Claude Fable 5: the cost of frontier AI

What Anthropic actually shipped

Two models landed on June 9. Claude Fable 5 (model ID claude-fable-5) is generally available on the Claude API, AWS, Amazon Bedrock, Vertex AI, and Microsoft Foundry. Claude Mythos 5 (claude-mythos-5) is the same underlying model with some safeguards lifted, offered only in limited availability through Project Glasswing, Anthropic's collaboration with the US government. Both are what Anthropic calls “Mythos-class,” a tier it positions above the Opus class.

The published specs: a 1M-token context window, 128k tokens of max output, adaptive thinking always on, and vision. One detail that matters for cost: Fable 5 uses the tokenizer introduced with Opus 4.7, so the same text is roughly 30% more tokens than older models. Your context fills faster, and you pay for the difference.

What it can do

Credit where it is due: the capability claims are not incremental. Anthropic reports that in a 50-million-line Ruby codebase, Fable 5 performed a codebase-wide migration in a day that would otherwise have taken a whole team over two months by hand. That figure, attributed to Stripe, is the kind of thing that changes what a small team can attempt.

On vision, Anthropic calls it the new state of the art: it can pull precise numbers from detailed scientific figures and rebuild a web app's source code from screenshots alone. It posts the highest score of any model on Hebbia's finance benchmark. Internally, Anthropic says Mythos 5 accelerated parts of its drug-design work by around ten times, and that in blinded comparisons its scientists preferred the model's molecular-biology hypotheses about 80% of the time over Opus-class models.

“The capabilities of models like Fable 5 and Mythos 5 have the potential to do profound good for the world.”

That is Anthropic's framing, and on the evidence it is not hype. Fable 5 is the most capable model most developers can call today. The interesting question is not whether it is good. It is what it costs to use it as your default.

The price, and what it means for your bill

Fable 5 costs $10 per million input tokens and $50 per million output, exactly double Opus 4.8 at $5 and $25. Opus-class output rates are already what drove this month's billing stories: Uber capped its engineers at $1,500 a month per tool after burning a year of AI budget in four months, and GitHub Copilot's move to metered AI Credits produced single-request burns that emptied monthly pools in days. Fable 5 doubles the ceiling on all of it, and the 30%-heavier tokenizer pushes the real number higher still.

The rollout follows the familiar pattern. From June 9 to 22, Fable 5 is included at no extra cost in the Pro, Max, Team, and Enterprise plans. On June 23, Anthropic pulls it from those plans and it starts drawing usage credits. Try it free, then meter it. There is nothing wrong with that as a business model. The thing to notice is that it is a meter, and a meter you do not control scales with how hard you work.

The safety gating is real, and it is the provider's call

Fable 5 ships with classifiers for three high-risk domains: cybersecurity, biology and chemistry, and distillation (extracting the model's capabilities to train a competitor). When a request trips one, the response is handled by Claude Opus 4.8 instead of Fable 5. Anthropic says this happens in less than 5% of sessions, so more than 95% involve no fallback. The fully ungated version, Mythos 5, is not something you can buy: it goes to approved Project Glasswing partners and the US government.

This is a defensible safety posture, and not a knock on Anthropic. But it is a clean illustration of a structural fact about cloud models: the provider decides what the model will and will not do for you, and can route your request to a different model without asking. That is fine for most work. It is a real constraint if your work lives in one of the gated domains, or if you simply need to know which model answered.

How to use frontier AI without the lock-in

The takeaway is not “do not use Fable 5.” Use it. It is the best public model available right now. The takeaway is how you use it.

The trap is making a $50-per-million-output cloud model your default, billed through a platform meter you cannot predict. The alternative is BYOLLM: bring your own Anthropic API key, pay Anthropic directly at API rates with no platform markup or credit pool, and reach for Fable 5 when a problem is genuinely worth it. For the routine work, run a local model like Qwen3.6 or DeepSeek V4 via Ollama or LM Studio at zero inference cost. The frontier model becomes a tool you pick up when you need it, not a subscription you are locked into.

That is the model Bodega One Code is built on. It is a local-first desktop IDE with BYOLLM: point it at your Anthropic key for frontier work, or at a local model for everything else, and switch between them without restarting. There is no platform meter and no per-model paywall, and the app is free for everyone in the open beta. One thing Fable 5 cannot do is run offline. It is cloud-only, so for air-gapped or regulated work where code cannot leave the machine, a local model is not a preference, it is the only option.

Fable 5 is a genuine step forward for what AI can do. Just decide, deliberately, how much of your work should depend on a meter someone else controls.

Sources

• Anthropic, June 9, 2026: Claude Fable 5 and Claude Mythos 5 (official announcement: capability claims, safeguards, pricing, availability): anthropic.com/news/claude-fable-5-mythos-5
• Anthropic, Models overview (official docs: model IDs claude-fable-5 / claude-mythos-5, 1M context, 128k output, $10/$50 pricing, Opus 4.8 comparison): platform.claude.com/docs/en/about-claude/models/overview
• Anthropic, Project Glasswing (Mythos 5 limited availability, US government collaboration): anthropic.com/glasswing
• TechCrunch, June 9, 2026: Anthropic released Claude Fable 5, its most powerful model, publicly: techcrunch.com
• Bodega One Code, June 9, 2026: Even Uber hit the AI coding cost wall (the enterprise cost context this builds on): bodegaone.ai/blog/agentic-ai-cost-wall