Claude Fable pricing: API rates and plan access

Claude Fable 5.1 costs $10 per million input tokens and $50 per million output tokens on Anthropic's API, the same as Fable 5, with cache reads cut 75% to $0.25 per million. The whole 1M-token context window is billed at that rate, and the Batch API halves it to $5 and $25. On Claude's subscriptions, Fable isn't available on the free plan, Pro and standard Team seats pay for it with usage credits, and Max plans and premium seats can spend up to half their weekly limits on it at no extra cost. Per token, it costs 2.5 times as much as Claude Opus 5.5 and the same as OpenAI's GPT-6 Astra.

Key takeaways

• API: $10 input, $50 output, $0.25 cache reads, $12.50 or $20 cache writes (5-minute or 1-hour), per million tokens.
• Fable 5 vs 5.1: the same input and output prices; Fable 5.1's cache reads are a quarter of Fable 5's $1. Anthropic estimates about 25% savings on typical workloads and up to about 45% on highly agentic ones.
• Plans: not on Free; usage credits on Pro and standard seats; up to 50% of weekly limits on Max and premium seats.
• Per task: $17.18 for an average Terminal-Bench 4.0 task on Vals AI's run, against $13.20 for Opus 5.5 and $9.58 for GPT-6 Astra.

Sources: Anthropic's pricing page, What's new in Claude Fable 5.1, Fable 5.1 announcement and Fable on your plan, opened October 27, 2026.

How much does the Claude Fable API cost?

Per million tokens, from Anthropic's pricing page:

• Input: $10.
• Output: $50. Fable 5.1's thinking is always on, and thinking tokens are billed as output.
• Cache writes: $12.50 for a 5-minute cache, $20 for a 1-hour cache.
• Cache reads: $0.25, which is 0.025 times the input price. That's the lowest ratio of any Claude model: Opus 5.5 and Sonnet 5.5 pay 0.05 times, and older models 0.1 times.
• Batch API: half price, $5 input and $25 output, for requests that can wait.

There's no surcharge for long prompts: a 900,000-token request costs the same per token as a 9,000-token one. Choosing US-only inference adds 10% to every rate. Anthropic's fast mode covers Opus models, not Fable. Claude Mythos 5.1, the same model for verified organizations, has the same prices.

Fable 5 vs Fable 5.1: what changed in price?

Only cache reads. Fable 5 charges $1 per million for them and Fable 5.1 $0.25; everything else matches. Agent sessions re-read their cached context on every step, so Anthropic estimates Fable 5.1 costs about 25% less than Fable 5 on typical workloads and up to about 45% less on highly agentic work. Vals AI's runs show the same direction: an average Terminal-Bench 4.0 task cost $30.34 on Fable 5 and $17.18 on Fable 5.1.

Which Claude plans include Fable?

From Anthropic's help center:

• Free: not available.
• Pro ($20 a month, $17 billed annually): Fable is billed to pay-as-you-go usage credits from the start, not to the plan's limits.
• Max 5x and 20x ($100 and $200 a month): Fable can use up to 50% of the weekly limit at no extra cost. It uses those limits faster than other models, and once the Fable share is used you can continue on credits or switch models.
• Team: premium seats ($100 or $125 a month) work like Max; standard seats ($20 or $25) use credits.
• Enterprise: premium seats work like Max and standard seats use credits; usage-based Enterprise plans pay API rates.

An earlier promotion let Pro and standard seats use part of their limits on Fable 5; it ended on July 19, 2026, and never covered Fable 5.1. More on plan limits in Claude Max vs Pro and Claude Code usage limits.

How does Fable's price compare?

Per million input and output tokens:

• Claude Opus 5.5: $4 and $20. Anthropic says it performs at Fable 5.1's level on most work; see Fable 5.1 vs Opus 5.5.
• Claude Sonnet 5.5: $2 and $10.
• GPT-6 Astra: $10 and $50, the same as Fable, but with $1 cache reads and double input pricing (1.5x output) on prompts over 272,000 tokens. See Fable 5.1 vs GPT-6 Astra.
• Kimi K3.1 Pro: $2 and $10, open weights promised by November 4. See Kimi K3 vs Claude.

Price per token isn't price per task. On Vals AI's Terminal-Bench 4.0 run, updated October 26, 2026, an average task cost $17.18 on Fable 5.1, $13.20 on Opus 5.5, $9.58 on GPT-6 Astra and $4.57 on Kimi K3.1 Pro.

How do you keep Fable costs down?

• Lower the effort. Claude Code runs Fable 5.1 at high effort by default and claude.ai at medium. On the API, Fable 5.1 can change effort mid-conversation (a beta) without breaking the cache, so routine steps can run lower.
• Keep the history append-only. Editing earlier turns invalidates Fable 5.1's thinking blocks and the cache; Anthropic suggests mid-conversation system messages instead.
• Batch what can wait, at half price.
• Use Opus 5.5 first and switch to Fable only for the steps that need it, as Anthropic's docs suggest.

Refusals are worth budgeting for: since September 24, 2026, some refusals that arrive before any output are billed. Anthropic's fallback credit refunds the cache cost of retrying on another model.

Can you use Fable without a Claude subscription?

Yes, through the API with prepaid credits, or through Amazon Bedrock, Google Cloud and Microsoft Foundry. Other tools resell it: Cursor and GitHub Copilot both charge Anthropic's $10 and $50. In Poly (usepoly.co), Fable 5.1 is one of 11 models on the paid plans (Pro is $20 a month, Max $50), and each person in a shared room picks their own model, so one teammate can use Fable while others use the free models.