Anthropic prices Claude Fable 5.2 at $8 per million input tokens and $40 per million output tokens, which it says is lower than Fable 5.1. The data terms are the part a procurement team has to read: customers not eligible for zero data retention get 30-day retention for safety monitoring, and AWS adds that Amazon staff review flagged traffic inside the AWS boundary.
What Anthropic announced
Anthropic’s launch post describes Claude Fable 5.2 and Claude Mythos 5.2 as the same underlying model. Fable carries cybersecurity and biology safeguards, which lets Anthropic offer it to all users. Mythos goes to organizations accepted into the company’s Life Sciences and Cyber Verification Programs. The post says both models generate output up to 30% faster than the 5.1 versions and are available at $8 and $40 per million input and output tokens. An optional fast mode generates up to 2.5 times faster and costs $16 and $80.
Anthropic’s own benchmark table puts Fable 5.2 at 49.1% on FrontierCode 2.0 at Xhigh effort, against 45.5% for Claude Opus 5.5, 42.8% for GPT-6 Astra and 41.1% for Gemini 4 Carbon. At High effort, Anthropic reports Fable 5.2 scoring 45.3%, level with Opus 5.5 at its maximum setting, for roughly half the cost per task. These are Anthropic’s measurements, and the post points readers to the system card for method. For most coding and knowledge work, Anthropic itself recommends starting with Opus 5.5 and moving up to Fable or Mythos only as the task requires.
The data terms behind the price
Anthropic explains that misuse of powerful models is often visible only across a window of requests, so Fable 5.2 relies on safeguards that watch usage patterns. Its answer for privacy-sensitive customers is Enterprise Frontier Safeguards, which runs the monitoring on prompts and responses stored in the customer’s own infrastructure and so works with zero data retention. The feature is in private beta with Bridgewater, Salesforce and Stripe. Eligible customers can keep using the models with zero data retention before adopting it. For everyone else, Anthropic states there is 30-day retention for safety monitoring.
AWS spells out what that means on its platform. In its Amazon Bedrock announcement, AWS says Anthropic designates Fable 5.2 a Covered Model, so by default its use requires retention for up to 30 days and human safety review, performed by Amazon personnel within the AWS boundary. An account has to set the aws_review data retention mode before it can invoke the model. AWS says eligible customers can use Fable 5.2 with zero data retention until Enterprise Frontier Safeguards is available, and that this is available for internal use through at least June 30, 2027.
What it means for the buying committee
The list price is the smaller decision here. A contract that already promises zero retention for prompts will not cover Fable 5.2 automatically. Legal and security teams need to ask the vendor which of the two paths applies to their account, and when the zero-retention exception ends for external use, since AWS gives a date only for internal use. Anthropic does not define who is eligible in the launch post, so the account team is the place to get that answer in writing.
Three questions belong on the checklist before a Fable 5.2 workload moves into production:
- Which retention path does the account sit on, and who at the cloud provider or Anthropic can read flagged prompts and responses during human review?
- If the account relies on zero retention today, what happens on June 30, 2027 and after Enterprise Frontier Safeguards leaves private beta?
- Does the application handle declined requests itself? AWS says a declined cybersecurity query can be retried on Opus 4.8 and a declined biology query on Opus 5, and that on Bedrock the retry is the caller’s job. A fallback credit token bills the retry’s cached input as cache reads.
The fallback point carries a cost angle. Output pricing at $40 per million tokens is only the bill for requests that the model answers. Applications that route declined requests to a second model carry two model IDs, two sets of evaluation results and, for regulated users, two data-handling reviews.
How it fits the pricing trend
The price cut sits alongside the trend in Ramp’s AI Index, where the blended price per million tokens across OpenAI and Anthropic fell 26.8% in four weeks. Our coverage of that index is in Ramp: Blended OpenAI and Anthropic Token Price Fell 26.8% in Four Weeks. A buyer comparing Fable 5.2 with a rival model needs to put retention, review and fallback terms next to the per-token rate, because the cheaper rate comes with those conditions.
Anthropic has also filed to go public. Its amended filing put third-quarter revenue at $17 billion, as we reported in Anthropic’s Amended IPO Filing Puts Q3 Revenue at $17 Billion. Teams negotiating multi-year commitments can ask now how the safeguard and retention terms will apply over the contract’s life.
Where it runs
Anthropic says Fable 5.2 is available today on all platforms, including Amazon Web Services, Google Cloud and Microsoft Azure. AWS says both models are on Amazon Bedrock through US Geo and Global cross-region inference profiles, and through Claude Platform on AWS. Fast mode is listed for Claude Code, the Claude Platform, Claude Models on Google Cloud and Microsoft Foundry. AWS says its prices are on the Amazon Bedrock pricing page, so a buyer on Bedrock should check the cloud rate against Anthropic’s list price rather than assume they match.
Source: Anthropic, Claude Fable 5.2 and Claude Mythos 5.2. Cloud terms from AWS Machine Learning Blog, Introducing Claude Fable 5.2 on AWS.
