brain/
← all entities
entitygenericartificial-intelligence

Claude Fable 5.1 / Mythos 5.1

Notes

Vintage: 2026-09. Primary evidence is Anthropic's official announcement and platform docs dated September 2026 (Fable 5.1 Released September 1, 2026), as hydrated in 2026-09-02-grok-com-ai-news-digest-2026-09-02-fable-5-1-astra-atlas-g20 and independently in the weekday X pass 2026-09-02-x-ai-overnight-fable-5-1-astra-critical-reward-seeker (fetch_method: x-mcp + product page). Capability and price claims are a snapshot of that issuer page, not a later model card. Demo videos on the @claudeai launch thread were not transcribed.

Claude Fable 5.1 / Mythos 5.1

One-line summary: Anthropic's September 1, 2026 Claude 5 point update — Fable 5.1 generally available, Mythos 5.1 the same model with different safeguards and trusted-access-only availability.

What it is

A Claude 5-generation point update on Fable 5. The official announcement says Fable 5.1 and Mythos 5.1 are the same model with different safeguards. Fable 5.1 is generally available. Mythos 5.1 is available only through trusted access programs; its safeguards are designed to support work in cybersecurity and the life sciences. Anthropic positions them as "the world's most advanced models for coding and knowledge work."

This page records issuer-hydrated claims from the Anthropic announcement and Fable 5.1 platform docs. The Grok.com digest that pointed at the release is a pointer, not grain. Aiwire's models index is an aggregator chip, not the issuer page.

Why it matters to this thread

Frontier model releases, pricing, and capability evaluations are in-scope. This is the first dated official Fable/Mythos 5.1 pair the thread has as a citable source. Distinct from claude-security (August 2026: Mythos 5 behind Enterprise scans, no separate model access) and from the June customer-by-customer direct Mythos 5 gate on government-gated-frontier-releases.

Key facts (from 2026-09-02-grok-com-ai-news-digest-2026-09-02-fable-5-1-astra-atlas-g20)

  • From the same source (Anthropic announcement, September 2026): Fable 5.1 is generally available; Mythos 5.1 is trusted-access only. Same model, different safeguards.
  • From the same source (Fable 5.1 platform docs): Released September 1, 2026.
  • From the same source (issuer pricing, confirmed on the Fable 5.1 overview): cache reads now cost 75% less, or $0.25 per million tokens. Typical workloads reduced by around 25% relative to Fable 5; complex coding / highly agentic tasks up to around 45%. Other token prices stay $10 / MTok input and $50 / MTok output (same as Fable 5).
  • From the same source (issuer Terminal-Bench table): Terminal-Bench 4.0 — Fable 5.1 55.8%, Mythos 5.1 60.9%, Fable 5 42.0%. Terminal-Bench-Science 0.1 — Fable 5.1 52.6% vs Fable 5 24.7%. Anthropic notes the Fable/Mythos gap on Terminal-Bench 4.0 reflects earlier, less precise cyber safeguards intervening. These are TB 4.0 / Science 0.1, not the official TB 2.0 harness-swing table on terminal-bench-harness-swing.
  • From the same source: the post answers customer feedback on price, data retention, and safeguards. Enterprise Frontier Safeguards (customer-controlled cloud storage) roll out in phases beginning later this fall; until then eligible customers can use Fable 5.1 with zero data retention.
  • From the same source: Mythos 5.1 is currently available to a set of US organizations. Two programs: Cyber Verification Program (Mythos-class access "in the near future") and Life Sciences Verification Program. Platform docs also route Mythos 5.1 to Project Glasswing participants.
  • From the same source (native X): @claudeai, 2:03 PM ET Sep 1 introducing both models; @AnthropicAI retweet.

Weekday X + product-page attach (2026-09-02)

  • From 2026-09-02-x-ai-overnight-fable-5-1-astra-critical-reward-seeker (official product page): API id claude-fable-5-1; list price otherwise matches Fable 5 at $10 / $50 per million input/output tokens with cache reads at $0.25 per million. Same-model / different-safeguards claim matches the grok.com digest hydration.
  • From the same source (product page; not in the grok.com digest grain): CursorBench 3.2.0 at 73.4% for Fable 5.1; "protein-binder design results for Mythos 5.1, and a Venus DEM scientific demo." No binder percentages or DEM scores in this pass — do not invent them. See claude-de-novo-protein-binders (August 2026 X claim; do not collapse this 5.1 mention into that experiment).
  • From the same source (official @claudeai X, 2026-09-01): Fable 5.1 "available everywhere today"; Mythos 5.1 stays in trusted-access programs for cyberdefenders and life scientists. Enterprise Frontier Safeguards for "customer-controlled privacy with misuse controls, rolling out in phases from this fall."
  • From the same source: Demo videos on the X launch thread were not transcribed; text + product page carry the claims. x_video: false on that source.

Adjacent price-axis color (2026-09-02/03) — not a Fable rewrite

Adjacent FLT formalization color (2026-09-04) — not a Fable rewrite

AA Intelligence Index v4.3 (2026-09-07) — not an Anthropic card

From 2026-09-07-artificial-analysis-intelligence-index-v4-3 — independent AA thread + fetched pages. Not a rewrite of issuer TB 4.0 55.8% or $10/$50. Prefer effort-labeled citations.

  • From the same source (announce): Index v4.3 53 (max with fallback), tied with gpt-6-astra (max).
  • From the same source (cost post): $7.63/Index-task vs Astra $3.26 (57% higher).
  • From the same source (TB4 thread): Terminal-Bench v4.0 52.0% (max with fallback). Page blurb: 55.1% (xhigh / default fallback). Effort mismatch, not one disputed number. Issuer table on this page remains 55.8% (Anthropic, unknown effort/harness). Do not flatten Fable TB4 to a single figure.
  • From the same source (AutomationBench thread): full-complete / no-violation 32.1% of workflows (max with fallback). Full treatment: automationbench-aa, artificial-analysis-intelligence-index.

AA Model Release pages (2026-09-09) — not an Anthropic card

From 2026-09-09-overnight-x-openai-astra-full-work-codex-rollout-aa-model — independent AA product/UI. Not a rewrite of issuer TB 4.0 55.8% or $10/$50. Full treatment: artificial-analysis-model-release-pages.

  • From the same source (@ArtificialAnlys, 2026-09-09 03:26:01Z): worked example (AA): Fable 47–53 Intelligence Index on 4.2–12.2 minutes per task vs gpt-6-astra 46–53 on 1.6–8.2 minutes.
  • From the same source (releases listing): Claude Fable 5.1 (Anthropic · Sep 2026 · 5 variants; Intelligence 47–53, cost/task $2.37–$7.63, 1M context). $7.63 high end matches the already-filed v4.3 max-with-fallback print — not a new max.

Adjacent alignment-assessment sim color (2026-09-09 afternoon) — not a Fable rewrite

UK AISI Mythos 5.1 access reports (2026-09-09/10) — not a Fable rewrite

Adjacent Sep 10 threat-intel misuse-detection color — not a Fable rewrite

What this source does not establish

  • Recap "restricted to vetted partners in cybersecurity and life sciences" is flattened. Issuer: trusted access programs; CVP Mythos-class access is "in the near future"; currently a set of US organizations.
  • Not a rewrite of claude-security. That page is Mythos 5 behind Enterprise scans (August 2026 @claudeai). This is a September 1 5.1 model release with a different access model for Mythos 5.1.
  • Not a rewrite of the June government release gate. Whether the June customer-by-customer direct-model gate still applies is not answered here.
  • Aiwire models section is aggregator index, not issuer.
  • No stock ticker, 8-K, or valuation print.

Open questions

Sources

Related

Referenced by
Concepts
AI Coding BenchmarksAnthropic alignment assessment of four cyber-eval unauthorized-access incidentsAnthropic illicit-distillation attributions (September 2026)Anthropic September 2026 threat intelligence reportArtificial Analysis Intelligence IndexArtificial Analysis Model Release pagesAutomationBench-AAChinese open-weight models approach frontier parityClaude de novo protein-binder design (Anthropic X claim)Claude formalizes Fermat's Last Theorem in Lean (Anthropic, Sep 4, 2026)Frontier intelligence is a perishable assetLLMs are a commodity; the moat is proprietary data + workflow integrationOpen vs closed source model economicsOpenAI Astra crossed Preparedness Framework "Critical" cyberTerminal-Bench harness swingThe US government is now in the frontier-model release loopUK AISI Mythos 5.1 pre-release access questions
Entities
Addy OsmaniAnthropicClaude CodeClaude Security (Mythos 5 scans)DeepSeek-V4.1-FlashFairwindGemini 3.8 Flash and 3.8 Flash CyberGPT-6 AstraLiam ByrneMoonshot AI (Kimi)Qwen3.8-Max-0902TypeSafe System One and JevWeatherNext 3