Anthropic cuts Claude Fable prices 20% and claims coding leadership with new speed features

Anthropic unveiled Claude Fable 5.2 and Mythos 5.2 this week, marking the latest iteration of its premium model pair. Both versions arrive with a 20 percent reduction in token pricing and a new acceleration feature designed to speed up response times.

Fable 5.2 rolls out immediately across all major cloud platforms—Amazon Web Services, Google Cloud, and Microsoft Azure. Mythos 5.2 access, by contrast, remains restricted to organizations enrolled in Anthropic's Cyber and Life Sciences verification programs. The two models share the same foundation but diverge in their safety configurations, with Fable enforcing stricter guardrails around cybersecurity and biology applications.

Notably, Anthropic did not bump these models to version 5.5, even though its Claude Opus, Sonnet, and Haiku lines now sit at that number. The company continues to position Opus 5.5 as its recommendation for most coding and knowledge work, while positioning Fable 5.2 for users grappling with "the most daunting technical challenges."

Speed and performance gains

Both Fable 5.2 and Mythos 5.2 generate output roughly 30 percent faster than their 5.1 predecessors. For users demanding even quicker responses, a new Fast mode accelerates token generation by up to 2.5 times. This mode is available through Claude Code, the Claude Platform, Claude Models on Google Cloud, and Anthropic-hosted Microsoft Foundry.

Pricing structure

Fable's token rates have dropped from $10/$50 per million input/output tokens to $8/$40, with Mythos 5.2 receiving identical pricing. This puts Anthropic 20 percent below OpenAI's GPT-6 Astra, which maintains standard rates of $10/$50 for prompts up to 272,000 input tokens. Fast mode carries a steeper cost: $16 per million input tokens and $80 per million output tokens—double the standard rates.

Even after the reduction, Fable's standard rates remain twice as high as Opus 5.5's pricing. However, Anthropic's latest evaluations suggest this premium does not necessarily translate to higher per-task expenses.

Coding benchmark results

On the FrontierCode 2.0 benchmark, Fable 5.2 achieved 49.1 percent, substantially ahead of Fable 5.1's 32.4 percent, Opus 5.5's 45.5 percent, and GPT-6 Astra's 42.8 percent. On FrontierSWE v2, Fable 5.2 scored 69.9 percent, edging out Astra at 65.5 percent and Opus at 62.3 percent.

A cost comparison with Opus reveals additional nuance. In early FrontierCode 2.0 testing, Fable 5.2 running at High effort matched Opus 5.5 at Max effort for roughly half the per-task cost. Medium effort is now the default setting for Fable 5.2 in Claude applications and Claude Code.

The results are not uniformly in Fable's favor. Google's Gemini 4 Carbon leads Fable on AutomationBench, which evaluates business workflows, scoring 53.9 percent versus Fable's 51.4 percent. Opus maintains a narrow advantage on the Elo measure of AA-Briefcase v1.1, a knowledge-work evaluation.

Agent coordination and prototype work

Anthropic describes Fable as capable of serving as an adviser that smaller models consult on difficult problems, or as an orchestrator delegating simpler tasks to them. In one internal test, an engineer tasked Fable 5.2 with building a prototype operating system from scratch. The model coordinated over 1,000 subagents across eight days and produced a bootable OS complete with drivers, persistent storage, a graphical interface, and an installer.

Safety blocks and guardrails

Anthropic maintains that Fable 5.2's cyber and biology safeguards remain as stringent as those in version 5.1. Early testing across Claude.ai, Claude Code, and Cowork showed approximately 55 percent fewer blocks overall, including roughly 60 percent fewer cybersecurity blocks and 40 percent fewer biology blocks.

Fewer blocks do not equate to broader permissions. Fable still prohibits high-risk cyber activities such as penetration testing and exploit generation, along with most research biology tasks. When a request is blocked, users receive a response from Opus 4.8 for cybersecurity matters or Opus 5 for biology.

Mythos access remains gated, but enterprise security teams have an alternative route: Claude Enterprise customers can now access Claude Security, which includes Mythos 5.2, without requiring membership in the Cyber Verification Program.

Anthropic deliberately constrained certain AI-development capabilities during training, including accelerator kernel optimization and work on large language model architectures. In these areas, performance matches Opus 5.5's level. The company states these limits do not affect agentic coding, hardware development, or traditional machine learning research.

For API developers, Anthropic is rolling out preserved thinking to all accounts. This feature prevents modifications to Claude's prior context that could extract its reasoning for distillation purposes.

Enterprise safeguards move to private beta

Enterprise Frontier Safeguards, first announced in September, is entering private beta with early customers including Bridgewater, Salesforce, and Stripe. The system runs automated safety monitoring on prompts and responses stored within customers' own infrastructure, enabling compatibility with zero data retention policies.

Eligible customers can continue using Fable 5.2 and Mythos 5.2 with zero data retention before implementing the system. For other customers, Anthropic retains data for 30 days to support safety monitoring.

Scientific research applications

Anthropic contends that Fable's coding strengths extend to scientific research, since much of that work relies on software for data analysis, molecular modeling, component design, and instrument control. The company reports that Fable 5.2 resolved two open theoretical-physics problems involving black holes and holographic duality.

One problem involved completing a 2003 calculation to determine whether black-hole formation remains an abrupt transition in a small-universe quantum description. The other established the limit of a technique used to rule out candidate theories of quantum gravity in a simplified universe.

Regarding the formalization of Fermat's Last Theorem, Anthropic clarifies that a team of 1,000 Fable 5.2 agents fully formalized the proof in Lean in 30 hours with minimal human intervention. Formalization translates a proof into a language that a computer can verify step by step—distinct from discovering a new proof. The agents received a list of intermediate theorem statements to track progress but no mathematical references. An earlier Claude research version required 11 days with a purpose-built harness and human guidance. Anthropic's comparison ran a single instance per model, with Opus 5.5 reaching 26 percent after 18 hours and making little additional progress by 30 hours.

Anthropic also evaluated Fable 5.2 on hardware design in collaboration with circuit-board company Diode, producing a voice-assistant hub, a USB speakerphone, and a handheld virtual pet. Claude selected components, designed the boards and enclosures, and wrote the software. Diode manufactured the boards, and its engineers tested and assembled the devices under Claude's direction. All three devices functioned without electrical modifications, according to Anthropic.