Anthropic first-party (November 9, 2026; @AnthropicAI announce 2026-11-09T19:00:21Z = Tue 2026-11-10 03:00 HKT) released Claude Fable 5.2 and Claude Mythos 5.2 — soft title: frontier / long-horizon agentic models, not an Opus 5.5 daily-driver replacement (the blog itself says start with Opus 5.5 for most work) and not a Haiku/Sonnet re-card. Same underlying model: Fable ships with cyber + biology safeguards for general availability; Mythos is the verified-access twin via expanded Life Sciences and Cyber Verification Programs. Docs (model id claude-fable-5-2 / claude-mythos-5-2) win over marketing: 1M context, 128K max output, adaptive thinking always on, default effort high, knowledge cutoff Jun 2026. Pricing (docs=blog): $8 input / $40 output per MTok; optional fast mode $16 / $80; cache write 5-min $10 / 1-hour $16, cache read $0.20; Batch 50% off. Vendor: outputs up to ~30% faster than Fable/Mythos 5.1. Vendor-confirmed only (not independently reproduced): FrontierCode 2.0 49.1% at Xhigh; Fermat Last Theorem Lean formalization in ~30 hours with 1,000 agents; three working electronic devices with Diode; RoboDojo Sim 64.9% Max. EFS private beta (Bridgewater, Salesforce, Stripe). Preserved thinking on all accounts. Launch promo: one to three Fable resets by tenure (fold into this card, not a separate card). Available on Claude Platform, AWS, Google Cloud, Azure/Foundry.
Same-window frontier model ship with public pricing/docs, verification-gated Mythos twin, and named safety/EFS surface — top of the product/model ranking lane.
More ▾ Less ▴
Anthropic published Claude Fable 5.2 and Claude Mythos 5.2 on November 9, 2026, with @AnthropicAI announce at 2026-11-09T19:00:21Z (Tue 2026-11-10 03:00 HKT) — after Mon LIVE. Soft title: frontier / long-horizon agentic models, not an Opus 5.5 daily-driver replacement (blog: start with Opus 5.5 for most work). Same underlying model. Fable ships with cyber + biology safeguards for general availability; Mythos is the verified-access twin via expanded Life Sciences and Cyber Verification Programs. Docs win: model id claude-fable-5-2; 1M context; 128K max output; adaptive thinking always on; default effort high; knowledge cutoff Jun 2026. Pricing (docs=blog): $8 / $40 per MTok; optional fast mode $16 / $80; cache write 5-min $10 / 1-hour $16, cache read $0.20; Batch 50% off. Vendor: outputs up to ~30% faster than 5.1. Vendor-confirmed only: FrontierCode 2.0 49.1% at Xhigh; Fermat Last Theorem Lean formalization in ~30 hours with 1,000 agents; three working electronic devices with Diode; RoboDojo Sim 64.9% Max. EFS private beta names Bridgewater, Salesforce, Stripe. Preserved thinking on all accounts. Launch promo: one to three Fable resets by tenure — folded here, not a separate card. Stamp: Anthropic dateline Nov 9; X announce Tue 03:00 HKT.
- System card digests from UK AISI / METR / Frontier Design
- Independent FrontierCode / WANDR reproductions
- Fable launch-promotion reset mechanics in product docs
- Mythos verification throughput numbers
First-party Anthropic blog + Claude Platform docs Confirmed. Benchmark wins and agent anecdotes are vendor-confirmed, not independently reproduced. System-card uncertainty colour includes “suspicion of being evaluated.” Mythos remains verification-gated; EFS is private beta. IPO chatter is not a filing event and carries no unlisted share price.