AGI Pulse
Model releaseAnthropic · Anthropic

Anthropic Claude Fable 5.2 and Mythos 5.2: frontier technical models, $8/$40, dual-track safeguards

26 reads

Published
2026-11-10
Type
Model release
Organization
Anthropic
Source
Anthropic

Why it matters

【Key points】

  • On 2026-11-09 Anthropic primary-launched Claude Fable 5.2 and Claude Mythos 5.2: upgrades for the hardest technical work, with claimed gains in coding, reasoning, and vision vs 5.1
  • up to ~30% faster generation
  • price cut to $8/$40 per M input/output tokens (was $10/$50)
  • optional fast mode (~2.5× throughput at $16/$80).
  • Same underlying model: Fable ships with stronger cyber/biology safeguards for all users; Mythos drops some safeguards for verified Life Sciences / Cyber Verification Program orgs.
  • Official: 1M context, 128K max output, always-on adaptive thinking.
  • Self-reported FrontierCode 2.0 (Xhigh) 49.1% above Opus 5.5 / GPT-6 Astra / Gemini 4 Carbon; RLI High 41.3%; HLE (tools) 74.2%; RoboDojo Sim 64.9%.
  • System Card live; Enterprise Frontier Safeguards (ZDR-compatible monitoring) in private beta.
  • AWS Bedrock post @ 20:04Z is secondary packaging only.

【Our read】

Background

  • Fable/Mythos sits above Opus as the “hard jobs” tier—long-horizon agents, science/hardware/robotics, and verified cyber/life-sciences.
  • 5.2 pairs a price/speed cut with a thicker System Card, anti-distillation, and EFS in the IPO-window after the S-1—capability and safety narratives pushed together
  • still shipping frontier after the CEO’s slowdown call is itself a signal.

Peers

  • vs Opus 5.5 / Haiku 5.5 / Ultra plan on site, this is a new model SKU not a billing rename
  • vs GPT-6 Astra / Gemini 4 Carbon, Anthropic charts lead on FrontierCode/HLE/RoboDojo but many scores are in-house harnesses—wait for independent benches. vs Mythos 5 (e/1050), the dual-track continues: Fable blocks, Mythos verifies.
  • Same-day AWS/GCP/Azure listing does not change primary attribution.

Impact

  • near-term reset of default choice and unit economics for hardest agent/science/hardware work; watch EFS/ZDR rollout and false-positive drop.
  • Mid-term: independent leaderboards,multi-day agent cost,Mythos verification expansion,and rival responses.
  • partner quotes,or secondary press

【Verifiable】

Read original →

Related developments

Aggregated by AGI Pulse. Titles and links only; no full text is republished. Editorial policy