AI Tools
What does Claude Fable 5.2 cost per million tokens?
Claude Fable 5.2 pricing is $8 per million input tokens and $40 per million output tokens, 20% below Fable 5.1 at $10 and $50, and fast mode doubles both to $16 and $80 for up to 2.5x faster output. Every figure here comes from Anthropic's own pricing pages, read on November 10, 2026.
What matters
- Fable 5.2 lists at $8 input and $40 output per million tokens, a 20% cut from Fable 5.1, and the Batch API halves it again to $4 and $20.
- On a 3:1 input-to-output mix the blended rate is $16 per million tokens, double Opus 5.5 at $8, so the flagship only saves money when it finishes a job in fewer tokens.
- Anthropic's own FrontierCode 2.0 numbers put Fable 5.2 at High effort at 45.3%, level with Opus 5.5 at Max (45.5%) for roughly half the cost per task.
- Fast mode bills $16 and $80, cannot be combined with the Batch API, and works out 20% cheaper per second of generated output at 2.5x speed.
- Cache reads cost $0.20 per million against a $10 cache write, so prompt layout usually moves the invoice more than model choice does.
What Claude Fable 5.2 actually costs
Claude Fable 5.2 is the frontier model Anthropic released on November 9, 2026. It shares weights with Claude Mythos 5.2 and sits above Opus 5.5 in the lineup, aimed at long-horizon agentic and scientific work rather than everyday coding.
Per Anthropic's pricing page, the API rate is $8 per million input tokens, $40 per million output tokens, $0.20 to read from cache and $10 to write to it. The Batch API takes 50% off both token rates, to $4 and $20. US-only inference adds a 1.1x multiplier on input and output. The model ID is claude-fable-5-2, and Anthropic says it is available on all platforms, including AWS, Google Cloud and Microsoft Azure.
Compare that with the model it replaces. Fable 5.1 still appears under legacy pricing at $10 input and $50 output, a $0.25 cache read and a $12.50 cache write. The 5.2 cut is exactly 20% across input, output and cache, which suggests a pricing decision rather than a rounding artifact.
Two numbers the launch post leads with are worth separating from the price. Anthropic says 5.2 generates output up to 30% faster than 5.1 on the standard tier. Fast mode is a separate paid switch that pushes generation up to 2.5x and doubles the token rates. Neither changes what a token costs on the default tier.
The 20% cut does not make this a cheap model
Blend the rates at three input tokens per output token, the shape most agent traffic takes, and Fable 5.2 lands at $16 per million tokens. Fable 5.1 was $20. Opus 5.5, the model Anthropic now calls its strongest daily driver, blends to $8. Sonnet 5.5 blends to $4.
So the flagship is twice Opus 5.5 per token on both sides of the meter, and four times Sonnet 5.5 on output. If you swap it into a fixed workflow that uses the same token counts, the bill doubles. The case for Fable 5.2 therefore rests entirely on the tokens it does not spend.
Anthropic argues exactly that. On FrontierCode 2.0, a benchmark that scores whether a model's commits would be merged by open-source maintainers, Fable 5.2 at High effort reaches 45.3% against Opus 5.5's best of 45.5%, for roughly half the cost per task in Anthropic's measurement. At Xhigh effort it reaches 49.1%, the highest score Anthropic reports, against Fable 5.1's 32.4%.
That is a real argument, but it is a per-task argument, not a per-token one. The honest reading: a task that costs $0.40 on Opus 5.5 can cost $0.20 on Fable 5.2 if it triggers fewer retries and shorter reasoning chains. A task that both models solve in one pass costs twice as much on Fable.
Fast mode: double the price for 2.5x the speed
Fast mode is the second billing tier for Fable 5.2 and Mythos 5.2, at $16 input and $80 output per million tokens. Anthropic's platform docs state that the premium applies across the full context window, including requests above 200K input tokens, that prompt-caching multipliers stack on top of it, and that fast mode is not available with the Batch API.
The arithmetic is friendlier than the headline. You pay 2x for 2.5x token generation, which is 80 cents per second of output compared with a dollar on the standard tier: 20% less per second of wall-clock generation. For an agent that spends most of its time waiting on the model, that is a straight improvement. For a batch summarization job that runs overnight, it is pure surcharge.
Availability matters for procurement. Fast mode for Fable 5.2 ships on Claude Code, the first-party Claude API, Claude Models on Google Cloud in both the Anthropic-managed and Google-managed variants, and Microsoft Foundry for Anthropic-hosted deployments. Google Cloud sets its own prices for the Google-managed path. Mythos 5.2 fast mode exists on the API, but only for organizations inside Anthropic's verification programs.
Cache writes are the line item to watch
A cache read on Fable 5.2 costs $0.20 per million tokens, 2.5% of the $8 input rate. A cache write costs $10 per million, 1.25x the input rate. Anthropic's page notes that the caching prices reflect a 5-minute time-to-live.
Run the numbers on a realistic agent. A 100,000-token system prompt with stable tool definitions, called 1,000 times in a working session: read from cache, the input side bills 1,000 x 0.1M x $0.20, or $20. Send it uncached and the same traffic bills about $800. One cache write at the front costs $1. Same tokens, same model, 40x difference in input spend.
Now flip the pattern. A workflow that rewrites its system prompt every call, or that spaces calls more than five minutes apart, pays the $10 write every time and gets nothing back. Effective input cost then lands at $10 per million against the $8 list price, which erases most of the 20% cut. Prompt hygiene beats model selection for anyone running Fable at volume.
Which Claude plan includes Fable 5.2
Fable is the first model Anthropic has gated this way on consumer plans, so read the plan table before assuming access. Pro lists at $20 per month when billed monthly, or $17 with an annual subscription that bills $200 up front, and it shows Fable as paid with usage credits rather than included in the plan allowance.
Max at $100 per month includes Fable up to half the weekly limit and adds $100 per month in API credits toward the Claude Platform. Ultra at $200 per month raises weekly usage to 10x Pro and the credit line to $200.
During launch week Anthropic is adding Fable usage on top of that: one to three Fable resets depending on how long you have subscribed, with Fable counting against up to 50% of weekly limits for Pro and standard Team seats. Treat that as a promotion with an end date, not as plan capacity, because the standing pricing page still lists Pro as credit-based for Fable.
When Opus 5.5 or Sonnet 5.5 is the better buy
Anthropic's own recommendation on the launch page is to start with Opus 5.5 for most coding and knowledge work and move up to Fable only when the task is open-ended enough to need it. The prices explain why: Opus 5.5 is $4 input and $20 output, Sonnet 5.5 is $2 and $10, and Haiku 5.5 is $0.10 and $0.50 for prompts up to 100K tokens, doubling to $0.50 and $2.50 above that.
The capability gap is not uniform. On FrontierSWE v2, Fable 5.2 reports 69.9% against Opus 5.5 at 62.3% and Fable 5.1 at 56.3%. On OSWorld 2.1 computer use it is 89.3% against 87.2%. Terminal-Bench-Science 0.1 shows the widest split of the set: 74.6% for Fable 5.2 against 58.7% for Opus 5.5 and 52.6% for Fable 5.1. Elsewhere the lead vanishes: Gemini 4 Carbon scores 53.9% on AutomationBench against Fable's 51.4%, and Opus 5.5 rates 1800 Elo on AA-Briefcase against 1785.
Practical split, given those numbers: Sonnet 5.5 for high-volume coding and extraction, Opus 5.5 as the default driver, Fable 5.2 for migrations, long agent runs and problems where a wrong answer costs more than the tokens.
A budget rule for Fable 5.2
- Batch everything that can wait. At $4 and $20 per million tokens, asynchronous work on Fable 5.2 still costs twice standard Sonnet 5.5 at $2 and $10, but it lands 60% below Fable 5.1's old $10 and $50 list.
- Keep fast mode for interactive work. It is 20% cheaper per second of generation and 100% more expensive per token. Pay for it where a human is waiting.
- Stabilize the prefix before you tune the prompt. Cache reads at $0.20 against $10 writes mean the first 100K tokens should be identical across calls.
- Measure tokens per task, not tokens per call. The 20% cut only pays off if Fable 5.2's fewer retries outweigh its higher per-token price on your workload.
- Give Fable to the hard 10%. Route routine coding to Opus 5.5 at $4 and $20 and escalate only the tasks that fail twice.
The head-to-head against the previous generation is at Fable 5.1 vs Mythos 5.1, and the wider rate card across Anthropic's tiers sits in Claude API pricing. Plan-by-plan allowances are compared at Claude pricing.
At a glance
| Model | Input | Output | Cache read | Cache write | Blended 3:1 |
|---|---|---|---|---|---|
| Claude Fable 5.2 | $8.00 | $40.00 | $0.20 | $10.00 | $16.00 |
| Claude Fable 5.1 (legacy) | $10.00 | $50.00 | $0.25 | $12.50 | $20.00 |
| Claude Opus 5.5 | $4.00 | $20.00 | $0.20 | $5.00 | $8.00 |
| Claude Sonnet 5.5 | $2.00 | $10.00 | $0.10 | $2.50 | $4.00 |
| Claude Haiku 5.5 (under 100K) | $0.10 | $0.50 | $0.01 | $0.125 | $0.20 |
FAQ
Is Claude Fable 5.2 cheaper than Claude Opus 5.5?
Per token, no: Fable 5.2 costs $8 input and $40 output per million against Opus 5.5 at $4 and $20. Anthropic's per-task claim is different, scoring Fable 5.2 at High effort level with Opus 5.5 at Max effort on FrontierCode 2.0 (45.3% vs 45.5%) for roughly half the cost per task. Cheaper per finished job, twice the price per token.
Does the Claude Pro plan include Fable 5.2?
Pro at $20 per month lists Fable as paid with usage credits, not as included plan usage. Max at $100 per month includes Fable up to half the weekly limit and adds $100 in monthly API credits. Anthropic's launch-week promotion temporarily adds one to three Fable resets for existing paid subscribers.
What is fast mode and when does it pay off?
Fast mode is an optional tier at $16 input and $80 output per million tokens that speeds token generation up to 2.5x, available on Claude Code, the Claude API, Google Cloud and Microsoft Foundry. It costs 2x per token but generates 2.5x faster, so it is 20% cheaper per second of output and only makes sense where a person or an agent is waiting on the result. It cannot be combined with the Batch API.