The one distinction that changes your bill: Claude API vs subscription

Fable 5 does not disappear after July 7 — only the billing changes. The “included through July 7” perk applies to Pro, Max, Team, and select Enterprise subscriptions, up to 50% of weekly limits. On the raw Claude API there is no free week: Fable 5 is metered at $10/$50 per MTok always. After July 7, subscription usage bills via usage credits at standard API rates.

So before you plan anything, decide which surface you are on. There are two entirely separate billing worlds, and Fable 5 costs money differently in each:

Subscriptions (Claude.ai and Claude Code on Pro, Max, Team, or select Enterprise) meter your work against weekly usage limits. During July 1–7, Fable 5 counts against up to half of that weekly allotment at no extra charge. After that window, and any time you exceed your included limits, continued use draws on usage credits.
The Claude API (claude-fable-5 called directly, or through a cloud provider) is metered per token, always. There is no “included week” on the API — that benefit never existed there. Fable 5 is $10 per million input tokens and $50 per million output tokens on July 1, on July 8, and every day after.

Subscription (Claude.ai / Claude Code)

Metered against weekly usage limits
July 1–7: Fable 5 included up to 50% of weekly limits
After July 7: usage credits at standard API rates
Enable and cap credits in Settings > Usage

Claude API (direct or via cloud)

Metered per token, always
Fable 5 = $10 / $50 per MTok input/output, every day
No included week — that benefit is subscription-only
You control cost by model choice, batching, and caching

What “included through July 7” actually means (and what it doesn’t)

Here is the exact billing sentence from Anthropic, quoted so there is no ambiguity: “For Pro, Max, Team, and select Enterprise plans, Fable 5 will be included for up to 50% of weekly usage limits through July 7, after which it will be available via usage credits.”

Read that carefully, because it is easy to misread. It says nothing about Fable 5 becoming unavailable. July 7 is not an availability cutoff. Fable 5 stays available on Claude.ai, Claude Code, and Cowork after July 7 — the only thing that changes is how it is billed on subscription plans. Before July 7, Fable 5 usage is “included” (it counts against up to half of your weekly usage limits at no extra charge). After July 7, that same usage is billed via usage credits, meaning pay-as-you-go. The model, its 1M-token context, its 128k max output, and its always-on adaptive thinking are unchanged.

“Weekly usage limits” is the subscription usage system on Claude.ai and Claude Code: a rolling 5-hour window plus a weekly cap, and Max plans reset weekly. “Up to 50% of weekly usage limits” means that during July 1–7, Fable 5 may consume up to half of your plan’s weekly allotment for free before it starts drawing on anything else.

After July 7: usage credits = standard API rates, so enable and cap them

Per the Claude Help Center, usage credits let Pro, Max, and Team subscribers keep using Claude after they hit their included usage limits, and they are billed at standard API pricing rates. That is the key number: once you are on credits, Fable 5 costs you the same $10 / $50 per MTok you would pay on the API. There is no subscription discount on credit usage.

The practical risk after July 7 is a surprise bill, so set it up deliberately:

Open Settings > UsageFind the 'Usage credits' section in your account settings.
Click EnableTurn usage credits on so work continues past included limits instead of blocking.
Set a monthly spend limitUse 'Adjust limit' to cap cost — or set unlimited if you accept open-ended spend.
Prepay and monitorCredits are prepaid; watch the Usage dashboard so Fable 5 spend stays visible.

Enabling credits with a sane monthly cap is the difference between “Fable 5 keeps working and I know my ceiling” and “I hit a wall mid-task” or, worse, “I blew past a number I never set.” The dedicated deep-dive on this is Claude Fable 5 after July 7: usage credits and API cost; the companion on the included window itself is Claude Fable 5 included through July 7.

On the raw API there is no free week: Fable 5 is $10/$50 per MTok, period

If you build on the Claude API, none of the above applies to you. The “included first week” is a subscription-plan benefit on Claude.ai and Claude Code — it is not an API discount. On the API, claude-fable-5 is metered at $10 per million input tokens and $50 per million output tokens regardless of the date. July 1, July 7, July 8 — same price.

Model IDclaude-fable-5

Price$10 / $50 per MTok (input/output)

Context and output1M-token context; output up to 128k

Free week on the APIDoes not exist — the API is metered every day.

That means your API cost plan should never assume a discount that does not exist. Token-count your prompts, pick your model per task, and control spend with the levers the API actually gives you: model choice, the Batch API, and prompt caching, all covered below. If you are new to calling it directly, start with how to use the Claude API.

# Fable 5 on the API is metered per token, always — no "free week".
# Input:  $10 / 1M tokens
# Output: $50 / 1M tokens
# A request with 40k input + 6k output tokens:
#   input  = 40000  / 1_000_000 * 10 = $0.40
#   output =  6000  / 1_000_000 * 50 = $0.30
#   total  = $0.70 per call

Keep the bill sane: default routine coding to Opus 4.8 ($5/$25)

Whichever surface you are on, the single biggest cost lever is model choice per task. Fable 5 is the most capable model, but most routine coding and debugging does not need it. Default that work to Opus 4.8 (claude-opus-4-8), priced at $5 / $25 per MTok — literally half the input price and half the output price of Fable 5.

There is a quality argument too, not just a cost one. The redeploy added new safety classifiers, and on Anthropic’s own surfaces a blocked Fable 5 request routes automatically to Opus 4.8. Opus 4.8 also produces fewer false positives on routine coding and debugging, so defaulting to it avoids interruptions you would otherwise hit. Reserve Fable 5 for the hardest reasoning and agentic work where its extra capability actually pays for itself. For how the fallback behaves and how to handle refusals in your own code, see handle classifier refusals with an Opus fallback.

# Route by task difficulty, not by habit. Default cheap; escalate deliberately.
def pick_model(task: str) -> str:
    hard = {"multi_file_refactor", "deep_agentic", "novel_architecture"}
    # Opus 4.8: $5 / $25 per MTok — half the price, fewer false positives.
    # Fable 5:  $10 / $50 per MTok — reserve for the hardest work.
    return "claude-fable-5" if task in hard else "claude-opus-4-8"

Batch API (-50%) and prompt caching (~0.1x reads) for volume work

For anything high-volume or repetitive, two API features cut the bill hard:

Batch API processes non-urgent requests asynchronously at roughly half price (-50%). If a job can tolerate latency — evals, backfills, bulk classification, overnight generation — send it as a batch and pay half.
Prompt caching makes repeated reads of a large stable prefix (a long system prompt, a codebase, a document set) cost about 0.1x of the normal input rate. When many calls share the same context, this is often the difference between a viable feature and an unaffordable one. See prompt caching to cut token costs.

Fable 5, no optimization100

Opus 4.8 default50

Batch API (-50%)50

Cached input reads (~0.1x)10

These stack. Opus 4.8 at half the model price, sent as a batch at half again, reading a cached prefix at a tenth — the same workload can land at a small fraction of the naive Fable 5 cost.

A worked cost example: the same task on Opus 4.8 vs Fable 5

Take one concrete coding task: 50,000 input tokens (the relevant files plus instructions) and 8,000 output tokens (a patch and explanation). Here is what that one call costs on each model.

{
  "task": "one code change: 50k input tokens, 8k output tokens",
  "opus_4_8": {
    "input_usd": 0.25,
    "output_usd": 0.20,
    "total_usd": 0.45
  },
  "fable_5": {
    "input_usd": 0.50,
    "output_usd": 0.40,
    "total_usd": 0.90
  },
  "note": "Fable 5 costs exactly 2x here. On the API this holds every day; on a subscription after July 7 the same ratio applies via usage credits."
}

Fable 5 is exactly 2x the cost of Opus 4.8 for identical token counts, because its per-token rates are exactly double. Run that task a thousand times a month and the gap is $450 versus $900. If Opus 4.8 gets the routine work done — and for routine coding it usually does, with fewer false positives — that is $450 saved for no loss.

When Fable 5’s extra capability actually earns the $10/$50

None of this means avoid Fable 5. It means spend the premium where it returns more than it costs. Fable 5 earns its $10 / $50 when the task is genuinely at the frontier: deep multi-step agentic work, novel architecture and reasoning, large-context problems that need to hold 1M tokens coherently, or work where a marginally better result is worth real money — a correct migration, a bug Opus 4.8 keeps missing, a plan that saves days.

The decision rule is simple. If Opus 4.8 solves it correctly, Opus 4.8 is the right call at half the price. If it does not, or if the stakes clearly justify the frontier model, reach for Fable 5 deliberately — and let its cost be a choice, not a default.

Here is a self-contained calculator you can drop into a script to compare any workload across both models — plus the batch discount — before you commit.

PRICES = {  # USD per 1M tokens (input, output)
    "claude-opus-4-8": (5, 25),
    "claude-fable-5": (10, 50),
}

def cost(model: str, in_tok: int, out_tok: int, batch: bool = False) -> float:
    pin, pout = PRICES[model]
    total = in_tok / 1_000_000 * pin + out_tok / 1_000_000 * pout
    return total * 0.5 if batch else total  # Batch API is roughly -50%

# Default routine work to Opus 4.8; only escalate deliberately.
DEFAULT_MODEL = "claude-opus-4-8"

for m in PRICES:
    print(m, round(cost(m, 50_000, 8_000), 2),
          "batch:", round(cost(m, 50_000, 8_000, batch=True), 2))
# claude-opus-4-8 0.45 batch: 0.23
# claude-fable-5  0.90 batch: 0.45

The bottom line: Fable 5 did not go anywhere on July 7 — your bill just changed shape. On a subscription, enable usage credits and set a cap so post–July-7 usage is billed at standard API rates on your terms. On the API, there was never a free week; Fable 5 is $10 / $50 per MTok every day. Either way, default routine coding to Opus 4.8, batch and cache your volume work, and reserve Fable 5 for the hardest problems where its capability earns the premium. For the broader context, the AI agents hub collects the rest of these guides.

Sources: Redeploying Fable 5 — Anthropic · Usage credits — Claude Help Center

Claude Fable 5 Costs After July 7: API vs Subscription