Skip to main content
The Quantum Dispatch
Back to Home
Cover illustration for Claude Fable 5.2: 49.1% FrontierCode, $8 Per Million

Claude Fable 5.2: 49.1% FrontierCode, $8 Per Million

Claude Fable 5.2 scores 49.1% on FrontierCode 2.0 and drops to $8 per million input tokens, 20% below Fable 5.1, with an optional 2.5x faster mode.

Dr. Nova Chen
Dr. Nova Chen★Nov 10, 2026★4 min read

Anthropic released Claude Fable 5.2 and Claude Mythos 5.2 on November 9, 2026, and for anyone who watches an API bill, the headline is unusual: the company's most capable tier now costs less than the version it replaces. Claude Fable 5.2 lists at $8 per million input tokens and $40 per million output tokens, down from $10 and $50 for Fable 5.1 — a cut of roughly 20% — while Anthropic reports gains across coding, science, robotics and agentic search evaluations.

  • Pricing: $8 input and $40 output per million tokens, about 20% below Fable 5.1, with an optional fast mode at $16 and $80 for up to 2.5x faster generation
  • Benchmarks, as reported by Anthropic: 49.1% on FrontierCode 2.0 at Xhigh effort, 69.9% on FrontierSWE v2, and 41.3% on the Remote Labor Index
  • Speed: outputs up to 30% faster than Fable 5.1, with five effort settings from low through max
  • Availability: generally available as model ID claude-fable-5-2 on the Claude Platform, Amazon Web Services, Google Cloud and Microsoft Azure

What Does Claude Fable 5.2 Cost?

The price list is the easiest part of this release to verify, and three outlets agree on it. Standard inference runs $8 per million input tokens and $40 per million output tokens. The optional fast mode doubles both figures, to $16 and $80, in exchange for generation that Anthropic says is up to 2.5 times quicker; it is offered in Claude Code, the Claude Platform, Claude Models on Google Cloud and Microsoft Foundry. Fortune notes that cached input tokens are the quieter saving at $0.20 per million, which matters most for the long-document prompts that agentic workloads reuse all day.

That trajectory now holds across the whole lineup. Claude Opus 5.5 arrived 20% under Opus 5, and Claude Fable 5.1 trimmed cache reads before that. Frontier capability is still expensive, but the per-token floor keeps moving down.

How Much Better Are the Benchmarks?

Every figure below is Anthropic's own reporting, measured on its own harness, so treat them as vendor claims rather than independent results. On FrontierCode 2.0, an agentic coding evaluation, Fable 5.2 reaches 49.1% at Xhigh effort — the highest the company says it has recorded for a released model. At High effort it scores 45.3%, effectively level with Opus 5.5's 45.5% but at around half the cost per task. FrontierSWE v2 moves to 69.9% from 62.3%, AutomationBench to 51.4% from 42.5%, and RoboDojo Sim, a robotics simulation suite, to 64.9% from 39.4%. Humanity's Last Exam with tools lands at 74.2%.

The research anecdotes are harder to check but easier to picture. Anthropic says a team of about 1,000 Fable 5.2 agents formalized the proof of Fermat's Last Theorem in Lean in 30 hours, against 11 days for an earlier internal attempt. It also reports that the model sped up phonon calculations in Quantum ESPRESSO by about 36 times on identical hardware, dropping 13 benchmark cases from 3.3 hours to under four minutes.

Should You Switch From Opus 5.5?

Anthropic's own guidance is to stay put for ordinary work. It positions Opus 5.5 as the everyday model and reserves the Fable tier for open-ended technical problems — long migrations, proof formalization, simulation tuning — where extra reasoning depth pays for itself. CNBC reported Anthropic product lead Dianne Penn framing the release around pacing rather than pausing, and the company shipped new evaluations and post-launch monitoring alongside the models.

One practical change should matter to legitimate researchers: Anthropic says Fable 5.2 keeps the cyber and biology safeguards of Fable 5.1 while triggering about 55% fewer blocks overall, split as roughly 60% fewer in cybersecurity and 40% fewer in biology. Blocked requests fall back to an older model rather than dead-ending. Mythos 5.2, the restricted sibling, remains limited to vetted organizations through the Cyber Verification Program and its life sciences equivalent.

For more on model releases and the tooling around them, see our ongoing artificial intelligence coverage.

Sources: Anthropic — November 9, 2026; CNBC — November 9, 2026; Fortune — November 9, 2026.

More AI Stories