Claude Fable 5

Anthropic's June 9, 2026 frontier release. **Claude Fable 5** is a Mythos-class model at general availability (model id `claude-fable-5`, **$10/M input, $50/M output**); **Mythos 5** is the same underlying model with safeguards lifted, restricted to authorized cybersecurity (Project Glasswing) and biomedical-research partners under a trusted-access program. **Closed weights, API-only** — the structural opposite of the self-hostable open models that fill the rest of this section. State-of-the-art on nearly all tested benchmarks; Stripe reported it compressed a **50-million-line Ruby migration from two months to one day**, and Mythos-5 protein-design partners reported roughly **10× drug-design acceleration**. Per [Anthropic — Claude Fable 5 and Mythos 5](https://www.anthropic.com/news/claude-fable-5-mythos-5).

Fable 5 is the **frontier anchor** of the 2026 cost-capability split this index maps. It does not run next to your object store — it's a rented API — but it defines the top tier the open floor ([DeepSeek V4](/node/deepseek-v4) and peers) is measured against. Its $10/$50 price point is what makes the floor *economically decisive*: at roughly **23× the input and 57× the output cost** of open weights, the frontier is reserved for the rare hard autonomous job, while the high-volume inference that actually touches your buckets — tagging, classification, retrieval, lifecycle — runs on the floor. Tellingly, Fable 5's own classifiers route the easy **~95% of sessions down to Claude Opus 4.8**, reserving the full model for the hard minority — the same difficulty-routing the market is doing across two vendors, done inside one product.

The hard, rare, high-stakes work where a few points of capability decides feasibility — large autonomous code migrations, novel agentic builds, frontier reasoning over long context; explicitly **not** the commodity data-plane inference that belongs on the open floor.