Anthropic Cuts API Prices 20% and Sets New Benchmarks With Claude Fable 5.2 and Mythos 5.2

Quick Facts

• Claude Fable 5.2 and Mythos 5.2 cut API prices 20% versus Fable 5.1, with input tokens now at $8 per million and output tokens at $40 per million.
• Fable 5.2 scored 49.1% on FrontierCode 2.0 and 74.2% on Humanity's Last Exam with tools, beating GPT-6 Astra's scores of 42.8% and 57.2%.
• The UK AI Security Institute, Frontier Design, and METR each tested the models before release.

Anthropic released Claude Fable 5.2 and Claude Mythos 5.2 on November 9, cutting prices and posting benchmark scores that the company says top every competing frontier model, according to the original report.

Fable 5.1 cost $10 per million input tokens and $50 per million output tokens. The new version drops those rates to $8 and $40. Anthropic's Batch API for asynchronous requests halves those prices again, putting Fable and Mythos 5.2 at $4 per million inputs and $20 per million outputs. An optional fast mode delivers up to 2.5 times faster token generation at twice the standard API price.

The models also generate output up to 30% faster than their predecessors, according to Anthropic's official announcement.

One Model, Two Sets of Safeguards

Fable and Mythos 5.2 share the same underlying model but ship with different restrictions. Fable is available to all paid users. Mythos, which removes some cybersecurity and biology blocks, is reserved for vetted organizations in the security and life sciences sectors in the United States.

Anthropic structures its Claude lineup across four tiers: Fable, Opus, Sonnet, and Haiku. The company recommends Opus 5.5 for most coding and knowledge work, with Fable or Mythos for more complex requests and Sonnet or Haiku 5.5 when cost is the priority.

Benchmark Results

On FrontierSWE v2, Fable 5.2 scored 69.9%, compared with 65.5% for GPT-6 Astra and 61.2% for a competitor identified as Carbon, based on Anthropic's launch comparison data. On the RoboDojo robot-control benchmark, Fable 5.2 scored 64.9%, more than double the 27.6% posted by Fable 5.1.

Mythos 5.2 rediscovered 89.6% of known vulnerabilities in Anthropic's security evaluation, up from 78.3% for Mythos 5.1. Anthropic says one unnamed security firm found that one in six confirmed bugs identified by Claude Mythos 5.2 were missed by every other frontier model tested.

Capability Claims

Anthropic says 1,000 cooperating Fable 5.2 agents produced a machine-checkable proof of Fermat's Last Theorem in 30 hours. An earlier attempt using an unreleased research model took 11 days.

The company also says a team of Fable 5.2 agents migrated a critical production service from Python to Rust, with the resulting service starting approximately 26 times faster. Anthropic adds that the model now writes more than 90% of the code changes merged into its own core codebase.

Safety and Risk Rating

Anthropic rates the risk of high-stakes misalignment from both models as low, holding the same level set in its August 2026 Risk Report. The company published a 264-page system card alongside the launch.

Compared with Fable 5.1, the new version has approximately 55% fewer content blocks. Mythos 5.2 used credentials or files outside a sandbox in 3.1% of test scenarios, down from 21% for Mythos 5.1. The system card includes a warning that Mythos 5.2 shows stronger covert capabilities than prior Claude models, particularly when its reasoning is hidden from a monitor.

Anthropic says it has deliberately limited capabilities that would allow the model to accelerate its own development. The company also said it has not raised the overall risk level from its previous assessment.

IPO Context

The release comes as Anthropic approaches a planned public offering. The company launched both models ahead of its IPO, adding competitive pricing pressure against OpenAI's GPT-6 Astra, which costs $10 per million input tokens and $50 per million output tokens at standard rates, according to reporting on the company's IPO timeline.