Anthropic launches Claude Fable 5.2 with 20% price cut

Anthropic launched Claude Fable 5.2 and Claude Mythos 5.2 on Monday, saying the two models share an underlying architecture and are the company's most capable for demanding technical work. Fable 5.2 is available on Amazon Web Services, Google Cloud, and Microsoft Azure. Direct access to Mythos remains restricted to organizations enrolled in Anthropic's Cyber and Life Sciences verification programs. According to Anthropic, both models can generate output up to 30 percent faster than their 5.1 predecessors. Each model accepts up to 1 million tokens of context and can return a maximum of 128,000 tokens, MLQ.ai reported. Anthropic still points users toward Opus 5.5 for typical coding and knowledge work, positioning Fable 5.2 for the hardest technical problems.

Fable's standard rate dropped from $10 and $50 per million input and output tokens for Fable 5.1 to $8 and $40. Mythos 5.2 has the same rates. A fast mode doubles those rates to $16 and $80 and can generate tokens up to 2.5 times faster. Users can access it in Claude Code, on the Claude Platform, via Claude Models on Google Cloud, and through Microsoft Foundry hosted by Anthropic. OpenAI's GPT-6 Astra continues to list at $10 and $50 for prompts up to 272,000 input tokens, so Anthropic now undercuts it by 20 percent on those token rates, according to The New Stack.

Fable 5.2 reached 49.1 percent on Anthropic's FrontierCode 2.0 coding benchmark. Fable 5.1 scored 32.4 percent, Opus 5.5 scored 45.5 percent, and GPT-6 Astra scored 42.8 percent. Its 69.9 percent on FrontierSWE v2 beat Astra's 65.5 percent and Opus's 62.3 percent. Anthropic reported that during early FrontierCode 2.0 testing, Fable 5.2 set to high effort matched Opus 5.5 at maximum effort while costing about half as much per task. Anthropic has made Medium the default effort level in Claude apps and Claude Code.

Anthropic highlighted long-running agent teams. About 1,000 Fable 5.2 agents took roughly 30 hours to produce a complete formal proof of Fermat's Last Theorem in Lean with little human assistance, the company said. During an internal experiment, an engineer instructed Fable 5.2 to create a prototype operating system starting from an empty directory. Anthropic stated that the model managed over 1,000 subagents across eight days and yielded an OS able to boot on a laptop, with an installer, drivers, persistent storage, and a graphical interface. With circuit-board maker Diode, Fable 5.2 chose components, designed boards and enclosures, and wrote software for three products: a hub with voice-assistant functions, a USB speakerphone, and a handheld virtual pet, according to Anthropic.

Anthropic said Fable 5.2's cyber and biology safeguards match the strength of 5.1's, yet early tests in Claude.ai, Claude Code, and Cowork logged roughly 55 percent fewer blocks overall; cybersecurity blocks fell about 60 percent and biology blocks about 40 percent. When the model blocks a query, it responds with Opus 4.8 for cybersecurity or Opus 5 for biology. Anthropic also intentionally held back certain AI-development skills, including accelerator kernel optimization, so the models perform around Opus 5.5's level. Enterprise Frontier Safeguards, announced in September, entered private beta with Bridgewater, Salesforce, and Stripe as early customers. Automated safety monitoring operates on prompts and responses kept in customers' own systems, which allows zero data retention, Anthropic said.

Mythos 5.2 matches Fable 5.2's specifications and pricing but stays restricted. MLQ.ai said access is limited to organizations verified through programs like the Cyber Verification Program. Claude Enterprise customers can get Mythos 5.2 through the Claude Security product without that verification. For cloud deployments, Fable 5.2 has model IDs on Google Cloud's Claude Models, Microsoft Foundry, and the Claude Platform on AWS. Anthropic's model overview does not list an Amazon Bedrock ID for it.

According to MLQ.ai's summary, CNBC said this is Anthropic's first frontier launch since CEO Dario Amodei's September call for the industry to slow capability gains, and that an IPO could occur as soon as next week. Dianne Penn, who leads product management for research at Anthropic, told CNBC, "We called for pacing, not pausing".