How to Use Claude Fable 5 After the Included Window

Update — July 1, 2026: Fable 5 is back worldwide — restored on July 1 after the US export-control directive was withdrawn on June 30. The billing schedule this post describes has been reset: instead of the original June 23 switch, Pro/Max/Team plans now include Fable 5 in up to 50% of the weekly usage cap through July 7, 2026, after which it moves to usage credits ($10/$50 per million tokens). Live status: /is-fable-5-back · What changed: Claude Fable 5 is back.

Every guide written in June answered the same question: how to use Fable 5 before the window closes. This one answers the question you'll have on July 8 — because the window closing is not the model disappearing, and the difference between a smart setup and an expensive one is about ten minutes of planning.

Counting down the days? Our live status page tracks the window in real time.

What actually changes on July 8

One thing: Fable 5 stops being included in the flat price of paid Claude plans. Since the July 1 restoration, every Pro, Max, Team, and eligible Enterprise subscriber can select Fable 5 in the model picker at no extra cost, within up to 50% of the weekly usage cap. On July 8, that selector doesn't vanish — it starts drawing on usage credits instead. (The original plan was a June 23 switch; the June 12–July 1 suspension pushed the whole schedule back.)

Everything else stays: the API price doesn't change, the cloud channels don't change (Bedrock, Vertex, and Foundry are being re-enabled post-restoration), the 30-day retention policy doesn't change, and the safeguard fallback doesn't change — in fact the new post-restoration safety classifiers make the Opus 4.8 fallback slightly more common while Anthropic tunes false positives.

Your four routes, costed

1. Usage credits in the Claude apps — for light, occasional use

Stay on your existing plan, keep selecting Fable 5 when you need it, pay for what you use on top. Zero setup. This is the right answer if Fable 5 is your "hard problems only" model a few times a week. Watch the meter the first week — chat sessions with long context add up faster than people expect, because every turn re-reads the conversation.

2. The API — cheapest per token, if you use the levers

Direct API access bills at $10 per million input tokens and $50 per million output — double Opus 4.8 on input, and the highest output rate in Claude's lineup. The sticker price is not the real price, though:

• Prompt caching cuts cached input by 90% ($1/M). Stable system prompts and shared context should always be cached.
• Batch processing takes 50% off non-urgent jobs.
• Fable 5 is also unusually token-efficient — it tends to reach correct answers with less reasoning output than competitors, which partially offsets the rate.

Run your actual workload numbers in our pricing calculator — it has per-token rates, caching, and batch discounts built in.

3. Cloud channels — if your infrastructure already lives there

AWS Bedrock, Google Vertex AI, and Microsoft Foundry all serve Fable 5 at parity pricing. Pick this for procurement or data-locality reasons, not price. Remember the retention terms follow the model, not the channel.

4. The hybrid: Opus 4.8 by default, Fable 5 on demand

The pattern that wins on cost for most teams. Opus 4.8 remains a very strong model at half the price, and for short, well-specified tasks the quality gap is small. Where Fable 5 pulls away is long-horizon agentic work — big refactors, multi-step research, autonomous sessions. Route by task difficulty: cheap model by default, Fable 5 when the task earns it. (Our Cursor and Claude Code guide shows how to set per-task model switching.)

Should you pay at all? A 30-second decision tree

1. Did Fable 5 visibly outperform Opus 4.8 for you during the included window? If you can't name a task where it did, stay on your plan's included models and revisit later.
2. Is your Fable 5 use case agentic/long-horizon? If yes, the price premium usually pays for itself in fewer retries and less babysitting — go API + caching.
3. Is it occasional hard questions in chat? Usage credits, no setup.
4. Is it cost-sensitive bulk work? It was never a Fable 5 job — batch it on Opus 4.8 or smaller.

The honest takeaway

July 7 is a billing event, not a capability event. The model that tops SWE-bench Pro on July 7 is identical on July 8 — what changes is that you'll choose it deliberately instead of by default. Most users need a routing habit more than they need a bigger budget: Opus 4.8 for the routine 80%, Fable 5 for the hard 20%, caching on everything stable. Set that up before the window closes and the transition is a non-event.

Pricing and policy details from Anthropic's launch and restoration documentation, updated July 2, 2026. Estimate your post-window costs with the Fable 5 pricing calculator.