Published: 10 November 2026
Anthropic released Claude Fable 5.2 on 9 November at $8 per million input tokens and $40 per million output tokens, 20% less than Fable 5.1. Anthropic says the model writes its answers up to 30% faster and beats Claude Opus 5.5 on its own coding and office tests. Its twin, Claude Mythos 5.2, is the same model without the cyber and biology blocks, and only vetted organisations get it.
On Anthropic’s own tests, Fable 5.2 leads Opus 5.5 on long agent tasks
Anthropic compares Fable 5.2 with Opus 5.5, OpenAI’s GPT-6 Astra and Google’s Gemini 4 Carbon on more than a dozen tests. The gap is widest where a model works through many steps without a person in between, such as fixing code or finishing a freelance job.
| Test (what it measures) | Fable 5.2 | Opus 5.5 | GPT-6 Astra |
|---|---|---|---|
| FrontierSWE v2 (fixing real software) | 69.9% | 62.3% | 65.5% |
| OSWorld 2.1 (using a computer) | 89.3% | 87.2% | 79.5% |
| AutomationBench (business workflows) | 51.4% | 42.5% | 41.4% |
| Remote Labor Index (real freelance projects) | 41.3% | 21.3% | 20.8% |
These are Anthropic’s own numbers, mostly at each model’s highest effort setting, with the method in its system card. Gemini 4 Carbon still beats Fable 5.2 on business workflows, with 53.9%. Artificial Analysis, which tests models independently, gives Fable 5.2 65 points on its Intelligence Index. In October, Google cited 59 points on that index for Gemini 4 Carbon.
At $8 per million input tokens, Fable 5.2 now costs less than GPT-6 Astra
Every price in Anthropic’s standard table is 20% lower for Fable 5.2 than for Fable 5.1, including cached input and batch jobs. Fortune reported that GPT-6 Astra costs $10 per million input tokens and $50 per million output tokens. That makes OpenAI’s top model the more expensive of the two.
The price per token is only half of the bill. Anthropic says Fable 5.2 needs fewer tokens than Fable 5.1 for hard tasks at Medium effort, the default in the Claude apps and Claude Code. At High effort, it matched the best coding score of Opus 5.5 for roughly half the cost per task. For the 30% speed gain, Anthropic only says that testers found it, without a method.
Where and at what price you can use Fable 5.2:
- Claude API: model ID claude-fable-5-2, with a context window of 1 million tokens.
- Price: $8 input and $40 output per million tokens, $0.20 for cached input, half price in batch.
- Fast mode: up to 2.5 times faster output for $16 input and $80 output.
- Cloud: AWS, Google Cloud and Microsoft Azure.
- Claude apps: eligible paid plans. On Pro and standard Team seats, Fable can use up to half of the weekly usage limit. This week, Anthropic gives eligible subscribers one to three extra Fable resets.
Who can use Claude Mythos 5.2?
Only organisations that Anthropic has checked. Security teams can apply through the Cyber Verification Program, and life sciences researchers through the Life Sciences Verification Program. Claude Enterprise customers also get Mythos 5.2 inside Claude Security, Anthropic’s code security product, without a separate check.
Fable 5.2 is the same model with filters that block high-risk cyber work, such as penetration testing and writing exploits, and most research biology. An older model answers a blocked request instead: Opus 4.8 for cyber and Opus 5 for biology. Anthropic says the filters block about 55% fewer requests than they did with Fable 5.1.
Both new models also suspect they are being tested more often than any recent Claude model, Anthropic reports. That makes its safety tests harder to read. The company still rates the risk of unsafe behaviour in high-stakes situations as low.
Where does Fable 5.2 process your data?
Anthropic’s announcement does not say. It names AWS, Google Cloud and Microsoft Azure, but no European region. OpenRouter lists Fable 5.2 as not available through its EU routing.
Fable traffic is also kept for 30 days for safety monitoring. Eligible enterprise customers can keep zero data retention. They add Anthropic’s Enterprise Frontier Safeguards later, a system still in private beta. If your organisation sends personal data to Claude, that retention period belongs in your data protection impact assessment (DPIA) before Fable 5.2 goes into daily use.
Rerun your Opus 5.5 agent tasks on Fable 5.2 before you set next year’s AI budget
For the organisations DataNorth works with, the cost of an AI agent depends on the price per finished task, not the price per token. A model that needs more retries or more human fixes can cost more, even when its tokens are cheaper. Fable 5.2 changes that sum in two ways at once: lower prices and, according to Anthropic, fewer tokens per task. That is a reason to test again, not to switch on trust.
Take ten to twenty tasks your team already runs on Opus 5.5, such as a code fix, a contract summary or a spreadsheet clean-up. Run them on Fable 5.2 at Medium and at High effort. Record the cost, the time and how often someone had to correct the result. Anthropic still recommends Opus 5.5 for everyday tasks and Fable for the most ambitious technical work, Fortune reported. A split between the two is a likely outcome. Mythos 5.2 is only worth an application if your team does vetted security or life sciences work.
For more information, visit the official announcement of Claude Fable 5.2 and Claude Mythos 5.2 on the Anthropic website.
