Claude Fable 5 Pricing and API Access

Claude Fable 5 pricing is $10.00 per million input tokens, $1.00 per million cache-read tokens, and $50.00 per million output tokens at Anthropic's standard API rates. OmniaKey currently lists $2.00 input, $0.20 cache read, and $10.00 output per million tokens.

The exact OmniaKey model ID is claude-fable-5. Fable has a 1M-token context window and always uses adaptive thinking, but context capacity is not prepaid usage. Every token sent, read from cache, or generated is billed in its corresponding category.

Claude Fable 5 pricing at a glance

All rates below are USD per one million tokens.

The Claude Fable 5 model page owns the live platform quote, specifications, calculator, and request example. This article owns the worked pricing and access decision. Use the model catalog to confirm availability before deployment.

The basic estimate is:

cost = uncached_input_millions × input_rate
     + cache_read_millions × cache_read_rate
     + output_millions × output_rate

Cache writes, tools, taxes, regional multipliers, retries, and special service modes can add separate charges. The compact formula is useful only when those items are absent or measured separately.

What each Fable 5 charge means

Uncached input

Input includes instructions, conversation history, repository files, tool definitions, and other request context that is not billed as a cache hit. One million uncached input tokens cost $10.00 at Anthropic list price or $2.00 at the current OmniaKey rate before Fable returns anything.

Cache reads

Repeated eligible context reported as a cache hit costs $1.00 per million tokens at Anthropic or $0.20 through the current OmniaKey rate. A repeated-looking prompt is not automatically a cache hit. Request structure, cache creation, lifetime, and provider reporting determine the billed category.

Anthropic separately prices cache writes. Do not apply the $1.00 read rate to newly stored context, and do not assume all conversation history remains cacheable forever.

Output and adaptive thinking

Output is the expensive line item: $50.00 per million tokens at Anthropic or $10.00 at the current OmniaKey rate. Fable 5's adaptive thinking is always enabled. Reasoning tokens that the provider reports as output are billable even when the visible answer is short.

Set a clear outcome and acceptance test. Artificially forcing a tiny output limit can interrupt a valuable investigation, while vague instructions can produce expensive tool loops and rework.

Three reproducible Fable 5 cost examples

These calculations use the displayed token categories directly. They exclude cache-write charges, paid tools, taxes, retries, and regional modifiers.

Example 1: one difficult investigation

Assume one million uncached input tokens and 100,000 output tokens.

Anthropic: 1 × $10.00 + 0.1 × $50.00 = $15.00
OmniaKey:  1 × $2.00  + 0.1 × $10.00 = $3.00

The 1M context window means the request can hold that much context. It does not make the first million input tokens free.

Example 2: a long autonomous workload

Assume several calls total ten million uncached input tokens and two million output tokens.

Anthropic: 10 × $10.00 + 2 × $50.00 = $200.00
OmniaKey:  10 × $2.00  + 2 × $10.00 = $40.00

This is arithmetic, not a prediction of how many tokens a repository task needs. Measure a representative run before setting a recurring budget.

Example 3: repeated cached context

Assume ten million tokens are reported as cache reads and the model produces one million output tokens.

Anthropic: 10 × $1.00 + 1 × $50.00 = $60.00
OmniaKey:  10 × $0.20 + 1 × $10.00 = $12.00

This example applies only to provider-reported cache hits. Cache creation is a separate category and may change the first-call economics.

Fable 5 versus Opus 5 cost

At standard Anthropic rates, Fable's input, cache-read, and output prices are exactly twice the corresponding Claude Opus 5 rates. The current OmniaKey rates preserve the same two-to-one relationship.

Start with Opus when both models can plausibly pass. Fable earns its premium when a controlled evaluation shows it resolving unusually ambiguous, long-horizon, or tool-heavy work that Opus cannot finish reliably.

Do not compare only cost per token. Compare cost per accepted task:

accepted-task cost = successful-run charges
                   + failed reruns
                   + human correction time
                   + remediation risk

The Claude Code model guide owns the broader Fable-versus-Opus-versus-Sonnet routing decision. The Opus 5 review owns the evidence for Opus. The Claude Code pricing guide compares subscriptions with API billing.

How to access Claude Fable 5 through the API

1. Confirm claude-fable-5 appears in the live catalog for your route.
2. Create a separate, limited key on API tokens instead of reusing a broad production key.
3. Store the value in OMNIAKEY_API_KEY; never put the real key in source code.
4. Use the Anthropic-compatible base URL https://api.omniakey.com.
5. Send the exact model ID claude-fable-5.

The localized API quick start owns the full setup. A minimal direct request is:

curl https://api.omniakey.com/v1/messages \
  -H "x-api-key: $OMNIAKEY_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "content-type: application/json" \
  -d '{
    "model": "claude-fable-5",
    "max_tokens": 1024,
    "messages": [
      {"role": "user", "content": "Identify the highest-risk assumption in this migration plan."}
    ]
  }'

For Claude Code, Fable requires version 2.1.170 or later. Anthropic's own availability can depend on organization eligibility and usage credits. A configured gateway route has its own catalog and billing boundary, so verify the active provider instead of assuming one access rule applies everywhere.

Common access and billing errors

Invalid API key or HTTP 401

Make sure the environment variable is available in the shell that launches the client, the key has not been revoked, and there are no quotes or spaces copied into its value. Rotate an exposed key rather than debugging it in public output.

Unknown or unavailable model

Use the exact ID claude-fable-5. Check the live catalog and your client's configured base URL. An unavailable model should return an error; OmniaKey does not silently substitute another model.

A safety refusal arrives as HTTP 200

Fable 5 can decline requests selected by its safety classifiers. On Anthropic's Messages API, this is a successful HTTP 200 response with stop_reason: "refusal", not a transport failure. Handle that response explicitly. Anthropic documents fallback options and says a request refused before output is generated is not billed; do not retry blindly in an unbounded loop.

Fable is missing from Claude Code

Upgrade to Claude Code 2.1.170 or later. On Anthropic's direct API, the picker can also reflect organization access, zero-data-retention restrictions, or usage-credit eligibility. Behind a gateway, model names and availability come from that gateway.

The 1M context window appears to be included credit

It is not. 1M describes request capacity. Input and cache tokens that occupy the window still incur their listed charges.

The estimate is lower than the invoice

Check cache writes, output-billed reasoning, tool charges, retries, regional settings, and taxes. Reconcile provider usage records rather than estimating from visible response text.

Budget controls that preserve useful work

• create one key per person, environment, or automation role;
• set a small initial limit around one representative task;
• log the requested model and actual uncached, cached, and output tokens;
• alert on loops, repeated failures, and unexpected model IDs;
• raise a limit only after reviewing accepted-task cost;
• route ordinary work to Opus or Sonnet and reserve Fable for a defined escalation class.

A budget that is too low can stop Fable immediately before verification. A budget with no task boundary can fund an unproductive loop. Calibrate from successful runs and preserve enough headroom for normal variation.

Frequently asked questions

How much does Claude Fable 5 cost?

Anthropic lists $10.00 input, $1.00 cache read, and $50.00 output per million tokens. OmniaKey currently lists $2.00, $0.20, and $10.00 respectively, checked August 12, 2026.

What is the Claude Fable 5 API model ID?

Use claude-fable-5. Pinning the exact ID makes tests and budgets easier to reproduce than relying on a changing family alias.

Does Fable 5 include one million free tokens?

No. The 1M figure is context capacity, not a token allowance. Usage is still billed as uncached input, cache activity, and output.

Is Fable 5 always better than Opus 5?

No. Fable targets the hardest and longest work and costs twice as much per token. Opus is often the better economic starting point; use Fable when measured completion quality justifies the premium.

Can I use Fable 5 with Claude Code?

Yes, when your active provider exposes it. Claude Code must be version 2.1.170 or later. Direct Anthropic access may also depend on organization and usage-credit rules.

For the cross-provider model decision rather than pricing, read Claude Fable 5 vs GPT-5.6 Sol for coding.

Sources checked

• Anthropic: introducing Claude Fable 5 and Claude Mythos 5
• Anthropic API pricing
• Claude Code model configuration
• OmniaKey Claude Fable 5 live model page
• OmniaKey API quick start

Fact-checked August 12, 2026. Model access, platform rates, cache behavior, and product rules can change. Verify the linked first-party pages and live catalog before making a production purchase.