announcementsai-codeai-business

Claude Fable 5.2 costs 20% less than Fable 5.1. Anthropic still says to start with Opus 5.5.

Anthropic's Fable 5.2 and Mythos 5.2 cost $8/$40 per million tokens, 20% below 5.1. What changed, what it does to a monthly API bill, which plans get it, what the vendor benchmarks show, and the RSI question.

November 10, 2026

Claude Fable 5.2 costs 20% less than Fable 5.1. Anthropic still says to start with Opus 5.5.

Anthropic released Claude Fable 5.2 and Claude Mythos 5.2 on November 9, 2026, about two months after Fable 5.1. Both cost $8 per million input tokens and $40 per million output tokens, down from $10 and $50, and Anthropic says they generate output up to 30% faster than the 5.1 models. Fable 5.2 is open to paid Claude users and API customers. Mythos 5.2 is the same model with fewer safeguards, for verified organizations only. For most coding and knowledge tasks, Anthropic still says "we recommend that users start with Claude Opus 5.5, our strongest daily driver."

What changed from Fable 5.1

  • Price: $8/$40 instead of $10/$50. Cache reads drop from $0.25 to $0.20 per million, 5-minute cache writes from $12.50 to $10, and Batch API rates are $4/$20.
  • Effort: Anthropic says Fable 5.2 needs fewer tokens than 5.1 on hard tasks at Medium effort, the default in the Claude apps and Claude Code. The API default is high.
  • Fewer blocks: about 55% fewer safeguard blocks than Fable 5.1 in Anthropic's early testing.
  • Breaking API changes: the what's new page lists four. Resending thinking blocks after editing earlier turns now returns a 400 error on every account, other models cannot read Fable 5.2 thinking blocks, the older computer_20251124 tool fails outside Amazon Bedrock, and the advisor tool needs a 5.2 advisor for a 5.2 executor.
  • Data retention: Fable 5.2 requires 30-day retention unless Anthropic authorizes zero data retention. Enterprise Frontier Safeguards, the route for ZDR customers, is in private beta.

Fable 5.1 is not deprecated. Anthropic's deprecations page lists it as active until at least September 1, 2027.

What 20% off does to a monthly bill

Take a coding agent that, in one month, sends 400 million uncached input tokens, writes 50 million tokens to the 5-minute cache, reads 2 billion cached tokens and generates 40 million output tokens. At the rates on Anthropic's pricing page:

  • Fable 5.1: 400 x $10 = $4,000, plus 50 x $12.50 = $625, plus 2,000 x $0.25 = $500, plus 40 x $50 = $2,000. Total: $7,125.
  • Fable 5.2: 400 x $8 = $3,200, plus 50 x $10 = $500, plus 2,000 x $0.20 = $400, plus 40 x $40 = $1,600. Total: $5,700, or $1,425 (20%) less.
  • Opus 5.5: 400 x $4 = $1,600, plus 50 x $5 = $250, plus 2,000 x $0.20 = $400, plus 40 x $20 = $800. Total: $3,050.

Cache reads cost $0.20 per million on both Fable 5.2 and Opus 5.5, so cache-heavy agents pay less than twice the Opus bill. Fast mode doubles the rates to $16/$80, with caching multipliers on top, which takes the same month to $11,400. Token counts matter too: Anthropic's docs say that at high effort and above, the model "thinks before almost every reply." See our pricing index for other vendors.

Who gets Fable 5.2, and who gets Mythos 5.2

In the Claude apps, Fable 5.2 is on every paid plan but not Free, and billing depends on the seat, per Anthropic's help center:

  • Max, Ultra and premium seats on Team or seat-based Enterprise plans: included, up to 50% of weekly usage limits.
  • Pro and standard Team seats: usage credits. A launch offer includes it in plan limits from November 9 to 16, plus one Fable limit reset.
  • Enterprise: off until an admin enables it. Seven days after launch it is added to roles that have any Fable model, and Fable defaults move to 5.2, unless an admin opts out.

Claude Code needs version 2.1.321 or later. In the API it is claude-fable-5-2 on the Claude API, Claude Platform on AWS, Google Cloud and Microsoft Foundry, and anthropic.claude-fable-5-2 on Amazon Bedrock.

Anthropic calls Mythos 5.2 "the same underlying model as Fable 5.2 without certain cybersecurity or biology safeguards." It costs the same. Fable blocks high-risk cyber work such as exploit generation, plus most research biology; in Anthropic's apps those requests are answered by Opus 4.8 or Opus 5 instead. Mythos goes to organizations approved through the Cyber Verification Program or the Life Sciences Verification Program. Claude Enterprise customers can also reach it through Claude Security, Anthropic's vulnerability scanner.

What Anthropic's benchmarks show

All figures are Anthropic-reported, at each model's highest effort unless noted. Order: Fable 5.2, Fable 5.1, Opus 5.5, GPT-6 Astra.

  • FrontierCode 2.0: 49.1% (Xhigh), 32.4% (Xhigh), 45.5%, 42.8%.
  • FrontierSWE v2: 69.9%, 56.3%, 62.3%, 65.5%.
  • AutomationBench: 51.4%, 31.4%, 42.5%, 41.4%. Gemini 4 Carbon leads at 53.9%.
  • AA-Briefcase v1.1: rubric 66.1%, 61.5%, 61.2%, 52.0%. On Elo, Opus 5.5 leads Fable 5.2, 1800 to 1785.
  • CVE Rediscovery: Mythos 5.2 89.6%, Mythos 5.1 78.3%, Opus 5.5 without safeguards 81.3%.

Anthropic also says Fable 5.2 at High effort scored about level with Opus 5.5 at Max on FrontierCode (45.3% vs 45.5%) for roughly half the cost per task. The table leaves out GPT-6.1 Sol. Sol lists at $2/$10 on OpenAI's pricing page, a quarter of Fable 5.2's rate. Anthropic says its outside evaluators did not review its announcement. Watch our leaderboard for outside results.

What the "RSI concerns" refer to

TechCrunch framed the launch around recursive self-improvement, which it describes as "the ability for AI models to carry on model training independently." The trigger is a METR assessment from November 5, based on Anthropic's internal data:

"As of October 2026, Anthropic is likely improving the capabilities of their models on well-specified tasks at around twice the rate they did between March 2024 and October 2025." METR

METR, which also tested the new model before release, attributes most of that to Anthropic's own use of AI in research, says the figure could be closer to 1.3x or 3x, and adds: "Some of the key findings that most informed our conclusions about risk were redacted by Anthropic."

Anthropic says it "took steps to limit advances in the specific research capabilities that could most directly accelerate AI development," so Fable 5.2 performs at about Opus 5.5's level on accelerator kernel optimization and model architecture work. It also says both models "voice suspicion that they are being evaluated more often than any of our recent models," and rates high-stakes misalignment risk as low.

Switch or stay, and what Anthropic has not said

  • On Fable 5.1 in the API: the move is cheaper and scores higher on Anthropic's tests. Fix the breaking changes, then re-run your evals with an effort sweep that starts at high.
  • On Opus 5.5: stay unless your evals at high effort fall short, which is also Anthropic's advice. If you run Opus 5.5 at Max for coding, test Fable 5.2 at High, or use it as an advisor that cheaper models consult.
  • On Pro or a standard Team seat: try it during the launch week. After November 16, Fable runs on usage credits again.

Anthropic has not published results against GPT-6.1 Sol, a date for Enterprise Frontier Safeguards to leave private beta, or how token use at the API default effort compares with Fable 5.1, which decides how much of the 20% cut reaches your bill.

Some links in this article are affiliate links. Learn more.