Claude Fable 5 was free in your plan. Now it's $10/$50 a token
Anthropic is moving Fable 5 off subscription plans to metered credits at $10/$50 per million tokens — double Opus 4.8. The lesson: bundled AI gets repriced.
If your team leaned on Claude Fable 5 because it came bundled with your subscription, today is the day that stops. As of July 7, 2026, Fable 5 is the last day inside Pro, Max, Team, and Enterprise plans; starting July 8 it bills through separate, funded usage credits at $10 per million input tokens and $50 per million output — double the price of Opus 4.8. It's a small change with a big lesson for anyone wiring AI into their business: the "free" model in your plan is a pricing decision someone else can reverse.
What actually happened
Per Anthropic's pricing page, Claude Fable 5 lists at $10/MTok input and $50/MTok output — versus Opus 4.8 at $5/$25. That makes Fable 5 the priciest generally available model Anthropic sells, exactly 2x its own flagship. (It also uses a newer tokenizer that produces roughly 30% more tokens for the same text, so the effective cost gap is wider than the sticker.)
The shift is in how you pay. As reported, through July 7 Fable 5 draws from your normal subscription limits; after that, it only answers if you've enabled a metered usage-credit balance — no grace period. A model that felt free inside a flat monthly fee becomes a per-token line item overnight. This is the same Fable 5 that briefly vanished under export controls last month — twice now, the thing you depended on changed without your input.
Why it matters for your business
Bundled AI is a promotion, not a guarantee. Labs seed a premium model into subscription tiers to drive adoption, then move it to metered billing once teams depend on it. If you built a workflow that assumes a specific top-end model is "included," you now have a cost you didn't budget for — or a feature that silently stopped working because nobody enabled credits.
The fix isn't loyalty to a plan. It's designing so the model behind each task is swappable. Most work — drafting, tagging, summarizing, routine support — runs fine on a cheaper model like Haiku or Sonnet 5. Reserve the priciest model for the narrow slice of work that actually needs it, and make that choice a config value, not a hard-coded dependency. Do that, and the next time a lab reprices a bundled model, you change one line instead of eating a bill or breaking a workflow.
Key takeaways
- Claude Fable 5 leaves subscription plans after July 7, 2026 — it now bills through separate usage credits with no grace period
- Metered pricing is $10/$50 per million input/output tokens — double Opus 4.8, the highest for a generally available Anthropic model
- Its newer tokenizer produces ~30% more tokens per text, widening the effective cost gap beyond the sticker price
- The operator move: don't build on a bundled "free" model — keep the model per task swappable so repricing costs you one config change, not a rewrite
Depending on a model that just got repriced? We build AI systems where each task points at the right-cost model — cheap work on cheap models, premium reserved for where it earns it — all swappable without a rewrite. See how we build portable AI systems or estimate what routing could save you.
Sources: Anthropic Pricing, Tech Times.
- #ai-costs
- #claude
- #anthropic
- #vendor-lock-in
- #portability
Tommy Rush — Founder, Rush Commerce
Operator turned builder. 15+ years running operations — now shipping the systems businesses run on. More
Get The Rush Report weekly — one email, zero fluff.
Keep reading
Super Intelligence Force: the FTC chair is on it. Check your AI claims
Trump's new Super Intelligence Force AI task force includes FTC Chairman Andrew Ferguson. No new rules yet, but your AI marketing claims are the exposure.
Read itReflection Beam: 501B open weights, Apache 2.0, 23B active
Reflection's Beam open-weight model claims GLM-5.2-level reasoning at 3-4x less inference compute. Apache 2.0 weights land later in October. What to test.
Read it