Claude Fable 5 Pricing Explained: Cost Per Million Tokens + Real-World Usage (2026)

Claude Fable 5 Pricing Explained: Cost Per Million Tokens + Real-World Usage (2026)

Heads up: Fable 5 was suspended on June 12, 2026, after US export controls were applied to it, and Anthropic redeployed it on July 1 (Anthropic). Anthropic said Pro, Max, Team and select Enterprise plans would include it for up to 50% of weekly usage limits through July 7, with usage credits required after that. If you want a second option anyway, see the best Fable 5 alternatives, including the open-source orchestrator Maestro and Sakana Fugu.

Prices checked September 26, 2026 against Anthropic's pricing page.

Claude Fable 5 costs $10 per million input tokens and $50 per million output tokens. Claude Fable 5.1, released September 1, 2026, keeps the same $10 and $50 rates and cuts cache reads from $1 to $0.25 per million tokens. Claude Opus 5.5, released September 22, 2026, costs $4 input and $20 output, so Fable is 2.5× its per-token price. Fable is also 2× the per-token price of Opus 4.8 and roughly 3× Sonnet 4.6. But per-token pricing is the wrong unit when comparing models for real work. What matters is cost per finished task, and on that dimension Fable 5 is sometimes the cheapest model and up to 17× more expensive than Sonnet 4.6 for the same outcome.

This guide breaks down Fable 5's actual cost on real workloads, the three patterns that blow up your bill, and the rule of thumb that protects you from the most common spend mistake.

If you're still setting up access, see our day-zero setup guide.

---

The Headline Numbers

Every model in this table keeps the same 5:1 output-to-input ratio. The Batch API takes 50% off input and output on every model. Sonnet 5.5 (released September 28, 2026) keeps Sonnet 5's $2 and $10 but halves cache reads to $0.10. Haiku 5.5 (October 7, 2026) is the cheap end of the ladder at $0.10 and $0.50, rising to $0.50 and $2.50 for prompts over 100,000 tokens. Table updated October 23, 2026 from Anthropic's pricing page.

What else the pricing page says, as of September 26, 2026:

• Cache writes: Fable 5 and 5.1 charge $12.50 per million tokens for a 5-minute cache write and $20 for a 1-hour write. Opus 5.5 charges $5 and $8.
• No long-context surcharge: Claude 4.6 and later models (except Haiku 5.5) bill the full 1M-token context window at standard rates. A 900K-token request costs the same per token as a 9K-token request.
• US-only inference: setting inference_geo to "us" multiplies every token category by 1.1×. Global routing, the default, uses standard prices.
• Context and output: Fable 5, Fable 5.1 and Opus 5.5 all have a 1M-token context window and 128K max output tokens.

For every other model's rates in one place, see our AI model price index.

---

What Changed in Fable 5.1

Fable 5.1 is now the current Fable model. Anthropic lists Fable 5 as a legacy model that is still available. Per Anthropic's docs, here is what changed:

• Same input and output price. $10 and $50 per million tokens, same cache write prices, same 50% batch discount.
• Cache reads cost a quarter as much. $0.25 per million tokens on Fable 5.1 (2.5% of the input price) versus $1 on Fable 5 (10%). This is the only price that moved.
• Stronger long-running work. Anthropic says Fable 5.1 brings stronger long-running agentic coding, multistep research, and document, spreadsheet and slide work.
• Three breaking API changes if you already call Fable 5: forced tool use returns an error, earlier models can't read its thinking blocks, and editing earlier turns invalidates thinking blocks.

The model ID is claude-fable-5-1. For the full rundown, read our Fable 5.1 what's new guide.

Fable 5.1 vs Opus 5.5

Anthropic's models overview now tells you to start with Opus 5.5 for most workloads and to use Fable 5.1 for demanding reasoning and long-horizon agentic work, or when your evals on Opus 5.5 at higher effort still fall short. On list price that is a 2.5× gap on both input and output. See our Opus 5.5 release breakdown for what the new Opus changes.

---

Cost Per Run, Not Per Token

The mistake every team makes in the first week with a new model is comparing per-token prices and concluding "Fable is 2× more expensive." That's true on a per-token basis. It's misleading on a per-finished-task basis, because different models use different numbers of tokens to complete the same task.

One more wrinkle: Anthropic says Claude 4.7 and later models use a newer tokenizer that produces about 30% more tokens for the same text than Sonnet 4.6 and earlier. The same prompt can count as more tokens on Fable than on Sonnet 4.6.

Three real-world workloads, measured during day-zero testing on Fable 5. Fable 5.1 bills input and output at the same rates, so the uncached costs below are the same on 5.1:

Workload 1: Quick code review on a 200-line PR

Fable 5 uses 3× the output tokens of Sonnet 4.6 here: it writes more thorough reasoning and surfaces edge cases the smaller models miss. The output is better, but you're paying about 8× more for it. On a quick code review, that's hard to justify.

Workload 2: Build a full /pricing page (Next.js, tests, theming)

Fable 5 costs 17× more than Sonnet 4.6 here, but it also finishes the task. Sonnet's output needs another 2 hours of human work to ship. If your time is worth $50/hour, Fable 5 is cheaper than Sonnet 4.6 on this task ($4.18 vs $100 of cleanup time).

Workload 3: Multi-hour async agent run (build a CRUD app end-to-end)

For long async runs, Sonnet 4.6 isn't actually a competitor: it doesn't maintain coherence over hour-long sessions. The real comparison is Opus 4.8 vs Fable 5, and the gap closes considerably because Opus is already capable here. Fable's $23 vs Opus's $5 is a real premium, but Fable's output is closer to "merge-ready."

We haven't run these workloads on Opus 5.5, so we won't guess its token counts. At list price, the same 80,000 input and 200,000 output tokens would cost $4.32 on Opus 5.5, but a different model can use a different number of tokens for the same job.

---

The Three Patterns That Blow Up Your Bill

Pattern 1: Using Fable 5 for chat

A 30-minute interactive coding session with Fable 5 over the API, lots of small back-and-forth turns, can easily run 50K+ output tokens. At $50/M output, that's about $2.50 for a conversation Sonnet 4.6 would bill at about $0.75 for the same 50K output tokens.

Fix: Use a Sonnet model for interactive chat. Save Fable for one-shot tasks where you don't need to iterate.

Pattern 2: Forgetting prompt caching on long sessions

Long Claude Code sessions accumulate context: system prompt, tool definitions, file contents read in earlier turns. Without prompt caching, you pay the input price ($10/M) for every token sent on every turn. With caching, cached reads cost 10% of the input price on Fable 5 ($1/M) and 2.5% on Fable 5.1 ($0.25/M). Re-reading 1 million cached tokens costs $0.25 on Fable 5.1, $1 on Fable 5, and $10 with no cache at all.

On a 4-hour Fable 5 session, this can be the difference between $25 and $80. On Fable 5.1 the cached share of that bill shrinks further.

Fix: Claude Code manages prompt caching automatically unless you disable it (Claude Code docs). If you're calling the API directly, add "cache_control": {"type": "ephemeral"} to your system prompt and tool blocks. The API docs have the full pattern.

Pattern 3: No max_tokens cap on long generations

Fable 5 will happily write a 30,000-token response if you let it. On output at $50/M, a single uncapped run can hit $1.50 just on output. Most tasks don't need 30K tokens. You're paying for unnecessary verbosity.

Fix: Set max_tokens explicitly. For code: 4,096-8,192 is plenty for most tasks. For research synthesis: 8,192-16,384. For "build this whole feature": let it run, but watch the dashboard.

---

When Fable 5 Is the Cheapest Option

Counterintuitively, Fable 5 can be the cheapest model for a task when:

1. You'd otherwise hire a contractor. A 4-hour Fable 5 run at $25 is dramatically cheaper than 4 hours of a senior contractor at $150/hour ($600), a gap our AI team cost calculator makes concrete for your own headcount. Even if you only use Fable for the 20% of tasks that would warrant a contractor, the math works.
2. Re-do cost is high. Shipping a buggy feature costs more than a thorough Fable 5 run. If Fable's higher quality reduces re-do rate by 30%, the per-task premium pays back.
3. The task is exactly Fable's sweet spot. Long, async, well-framed, complex. Fable was designed for this. Cheaper models will iterate longer and use more total tokens to reach the same quality.

---

When Fable 5 Is the Most Expensive Mistake

1. Interactive chat or quick edits. Use a Sonnet model.
2. Tasks Opus already handles well. Single-file edits, simple refactors, bug fixes, documentation, code review on small PRs. You're paying 2× Opus 4.8's rate, or 2.5× Opus 5.5's, for marginal quality improvement.
3. Anything where you don't yet know what you want. Iteration is faster and cheaper on Sonnet/Opus. Use Fable when you've already clarified the brief.

---

A Monthly Budget Model

If you're trying to predict monthly spend, here's a reasonable starting point for a single developer:

These numbers are conservative for serious users and scale up with team size. The "production Claude Code team" line assumes 4-6 engineers using Claude Code daily on real work. The model mix was built before Opus 5.5 and Sonnet 5's standard pricing: both list below the Opus 4.8 and Sonnet 4.6 rates used here, so moving that share of work to them lowers the per-token cost of those columns. Fable 5.1 bills the Fable column at the same input and output rates.

For comparison, that $1,400/month for a 5-engineer team is about 9 hours of a senior contractor at $150/hour. If the AI saves each engineer 2 hours a month, the spend pays for itself.

---

How to Actually Control Spend

Three practical disciplines:

1. Default to a Sonnet model

Honestly, this is the biggest lever. Set a Sonnet model as your default in Claude Code (run /model inside a session, or start it with claude --model claude-sonnet-5-5). Sonnet 5.5 lists at $2/$10 per million tokens, below Sonnet 4.6's $3/$15. Switch up to Opus or Fable explicitly when the task warrants it.

2. Cap max_tokens per request

Every API call should have an explicit max_tokens. Pick the smallest value that fits your real output. For most coding: 4,096. For most chat: 1,024. You'd be surprised how often you don't need more.

3. Use prompt caching on long sessions

If you're running multi-hour Claude Code sessions or building an agent that runs over many turns, prompt caching cuts your input bill by 90% on repeated context with Fable 5, and by 97.5% with Fable 5.1. Claude Code manages it automatically unless you disable it; in your own API integrations, add the cache_control flag.

---

API vs Claude Subscriptions

A nuance worth knowing: Claude Pro and Max subscriptions include usage limits instead of per-token billing. Claude's pricing page lists Pro at $20/month billed monthly ($17/month on an annual plan) and Max from $100/month, with 5× or 20× more usage than Pro. Limits reset on a rolling five-hour session window. The practical implication: subscription users hit a usage limit rather than seeing a per-task charge.

If you start running Fable heavily on a subscription, expect to hit those limits much sooner than on lighter models. For predictable production workloads, the API is usually cheaper than scaling subscription seats. For everyday developer use, a subscription is simpler.

---

FAQ

How much does Claude Fable 5 cost per token?

Claude Fable 5 costs $10 per million input tokens and $50 per million output tokens, a 5:1 output-to-input ratio. That's 2x the per-token price of Claude Opus 4.8, roughly 3x Claude Sonnet 4.6, and 2.5x the new Claude Opus 5.5.

Is Fable 5 more expensive than Opus 4.8?

Yes, on a per-token basis Fable 5 costs 2x Opus 4.8 ($10/$50 vs $5/$25 per million tokens). Per finished task the gap varies: Fable 5 can end up cheaper than Opus on long async runs where Opus needs more iteration, or far more expensive on simple tasks Opus already handles well.

Is Fable 5.1 more expensive than Fable 5?

No. Fable 5.1 costs the same $10 per million input tokens and $50 per million output tokens as Fable 5, with the same cache write and batch prices. The only price change is cheaper cache reads: $0.25 per million tokens on Fable 5.1 versus $1 on Fable 5, so cache-heavy workloads cost less on 5.1.

How does Fable 5.1 pricing compare to Opus 5.5?

Claude Opus 5.5 costs $4 per million input tokens and $20 per million output tokens, so Fable 5.1 is 2.5x more per token on both. Cache reads are close: $0.20 on Opus 5.5 versus $0.25 on Fable 5.1. Anthropic recommends starting with Opus 5.5 and moving to Fable 5.1 for demanding reasoning and long-horizon agentic work.

How do I estimate my monthly Fable cost?

Multiply your monthly input tokens in millions by $10 and your output tokens in millions by $50, then add them. For example, 20 million input and 4 million output tokens costs $200 plus $200, so $400 a month. Cached reads bill at $0.25 per million on Fable 5.1, and the Batch API halves both rates.

How much does a typical Fable 5 task actually cost?

It depends on the workload, not the sticker price. A quick code review on a 200-line PR runs about $0.10 with Fable 5. Building a full Next.js page end-to-end runs closer to $4. A multi-hour async agent run building a CRUD app can cost around $24. See the workload tables above for the full cost-per-task breakdown against Sonnet 4.6 and Opus 4.8.

Is there a cheaper Fable 5 alternative?

Within Anthropic's lineup, Claude Opus 5.5 at $4/$20 per million tokens is the obvious step down, and Sonnet 5.5 at $2/$10 is cheaper still for everyday interactive coding and chat. For classification and routing, Haiku 5.5 lists at $0.10/$0.50. For teams that want an open-source option instead of paying per token at all, see our guides to Fable 5 alternatives, the Maestro orchestrator, and Sakana Fugu.

How much does Fable 5 cost per month for a team?

For a single heavy IDE user, expect around $80/month. For a 4-6 engineer team using Claude Code daily on production work (the "production Claude Code team" profile), budget roughly $1,400/month combined across Sonnet, Opus, and Fable 5 usage. See the full monthly budget model above for lighter usage profiles.

Can I reduce my Fable 5 bill without switching models?

Yes. Three habits cut spend the most: default to a Sonnet model for everyday work and reserve Fable for tasks that need it, cap max_tokens explicitly on every request instead of letting generations run long, and use prompt caching on long sessions, which cuts the repeated input-token bill by 90% on Fable 5 and 97.5% on Fable 5.1.

---

Bottom Line

• Per token, Fable 5 and Fable 5.1 cost $10/$50: 2× Opus 4.8, 2.5× Opus 5.5 and roughly 3× Sonnet 4.6.
• Fable 5.1 changes one price: cache reads drop to $0.25 per million tokens from $1.
• Per finished task, the multiplier varies 1×-17× depending on workload and quality requirements.
• Default to Sonnet, escalate to Opus 5.5 for hard tasks, and reach for Fable 5.1 when the assignment would otherwise warrant a senior contractor.
• Three habits matter: cap max_tokens, use prompt caching, and don't use Fable for interactive chat.

For the full picture of how to use the model, see our day-zero setup guide and the Fable 5 vs Opus 4.8 comparison. Migrating after the suspension? See our Claude Fable 5 alternatives.

---

Want Help Picking the Right Model for Your Production Workload?

Production Claude usage is full of subtle decisions: which model per task, where to add prompt caching, how to set max_tokens per endpoint, where to add fallbacks when usage limits hit.

AY Automate places senior AI engineers into your team from $60,000 a year, in 2 to 4 weeks, with a 90-day replacement guarantee, or if you need a system built rather than staffed, our AI agent development team ships production agents your engineers can own. We make the cost-optimization decisions so you can ship faster without the surprise bills. Our AI strategy consulting team reviews your model routing, caching setup, and token budget and returns a prioritized savings plan. Book a free 30-min strategy call. We'll look at your current spend and tell you where the biggest wins are.

Keep reading

• local LLM tools that cut your token bill to zero
• every free AI model and API, with real limits
• open-source agent frameworks to pair with it