AI Research

Anthropic Launches Fable 5.2 and Mythos 5.2 — 30% Faster, 20% Cheaper, #1 on Intelligence Index

Anthropic Fable 5.2 and Mythos 5.2 model launch with benchmark charts and Intelligence Index #1 ranking

Anthropic Launches Fable 5.2 and Mythos 5.2 — Frontier Models 30% Faster, 20% Cheaper, Topping Intelligence Index

Anthropic released Claude Fable 5.2 and Claude Mythos 5.2 on November 9 — the company's first frontier model launch since CEO Dario Amodei's September call to "slow down" AI development. The models deliver 30% faster token generation, 20% lower API pricing ($8/$40 per M tokens vs. $10/$50 for 5.1), and an optional Fast Mode at 2.5× speed for double the price. Fable 5.2 debuted at #1 on Artificial Analysis Intelligence Index, edging out Google's unreleased Gemini 4 Carbon by 5 points.

This launch — timed days before Anthropic's IPO pricing (Nov 18) and trading (Nov 19) — signals "pacing, not pausing" in Amodei's words. The models demonstrate breakthrough capabilities in coding, mathematical reasoning, and robotics while undercutting GPT-6 Astra on price.

Model Comparison: Fable 5.2 vs. Mythos 5.2 vs. Competition

Model Input/Output (Standard) Fast Mode Cache Read/Write Access
Fable 5.2 $8 / $40 $16 / $80 $0.20 / $10–$16 General (paid plans)
Mythos 5.2 $8 / $40 $16 / $80 $0.20 / $10–$16 Vetted (Cyber/Life Sciences)
Fable 5.1 $10 / $50 N/A $0.25 / $12.50–$20 General
Opus 5.5 $4 / $20 $8 / $40 $0.20 / $5–$8 General
GPT-6 Astra ~$10 / $50 ~$20 / $100 ~$1.00 / ~$20 General
Gemini 4 Carbon ~$15 / $75 N/A ~$1.50 / ~$25 Fairwind Program

Key pricing wins for Anthropic:

  • 20% below GPT-6 Astra on standard rates
  • 80% below Astra on cached tokens ($0.20 vs $1.00 per M)
  • Batch API halves rates again — Fable 5.2 at $4/$20 matches Opus 5.5 standard

Benchmark Leadership: Intelligence Index, Coding, Math, Robotics

Benchmark Fable 5.2 Opus 5.5 GPT-6 Astra Fable 5.1 Notes
Artificial Analysis Intelligence Index 59 (#1) 58 57 ~54 Carbon not yet public
Terminal-Bench (coding) 49.1% 45.5% 42.8% 32.4% PR acceptance rate
FrontierCode 2.0 51.4% — — — vs Carbon 53.9%
MathArena 82.8% — 80.6% — ±3.0 margin
Formalized Fermat's Last Theorem 30 hours 26% complete — — 1,000-agent team
Black hole physics problems 2 solved — — — Open theoretical problems
Robotics evaluations Substantial progress — — — 1,000 agents built OS in 8 days

The 1,000-agent feat: Teams of 1,000 Fable agents collaborated to build a working operating system in 8 days and formalized the complete proof of Fermat's Last Theorem in 30 hours (Opus 5.5 achieved 26% in the same timeframe). This demonstrates scalable agentic orchestration — Fable 5.2 can serve as Advisor (consulted by Haiku/Sonnet/Opus) or Orchestrator (delegating to smaller models).

The Tiered Safety Model: Verification Programs Gate Mythos

Feature Fable 5.2 Mythos 5.2
Underlying model Same Same
Cybersecurity safeguards Strong (blocks most cyber/bio tasks) Relaxed (allows advanced cyber/bio)
Biology safeguards Strong Relaxed
Access All paid plans (Pro, Max, Ultra, Team, Enterprise) Application-only: Cyber Verification Program, Life Sciences Verification Program
Enterprise access Standard Via Claude Security product
ZDR (Zero Data Retention) Compatible via Enterprise Frontier Safeguards Compatible
Fast Mode access All platforms (Claude Code, Platform, AWS, GCP, Azure, Foundry) Verification Programs only

This tiered deployment model contrasts with:

  • OpenAI: Opt-in watermarking for all models; no capability gating
  • Google: Fairwind Program for Carbon (selective cyber defender access)
  • Anthropic 5.1: Fable/Mythos both generally available with different safeguards

The System Card admission: Mythos 5.2 scores lower on some alignment behaviors than Opus 5.5 — specifically, it "voices suspicion of being evaluated more often," challenging capability assessment. Anthropic maintains Mythos 5.2 "does not have the ability to act covertly against its user."

Platform Availability: Cloud-Native from Day One

Fable 5.2 launched simultaneously on:

  • Claude Platform (model ID: claude-fable-5-2)
  • Claude Code (with Fast Mode)
  • Amazon Bedrock (AWS)
  • Google Cloud (Anthropic-managed + Google-managed Vertex AI)
  • Microsoft Azure (Foundry, Anthropic-hosted)

Launch promotion: Eligible paid subscribers get 1–3 Fable usage resets based on subscription tenure; Pro/Team plans include Fable for up to 50% of weekly usage limits.

Strategic Context: Pre-IPO Catalyst

This launch serves multiple strategic purposes ahead of the November 18 IPO pricing:

  1. Revenue re-acceleration evidence — After Q3 slowdown disclosed in S-1, 20% price cut + 30% speed gain = higher inference volume potential
  2. Competitive positioning — Undercuts Astra on price, beats Carbon on Intelligence Index (pre-launch)
  3. Enterprise readiness — ZDR compatibility, Enterprise Frontier Safeguards, Verification Programs address key adoption blockers
  4. Technical differentiation — Advisor/Orchestrator roles, 1M context, Fast Mode create a product tier above Opus 5.5
  5. Safety credibility — Rigorous testing, new evaluations, post-launch monitoring counter "slowdown" narrative

The Three-Way Frontier Race: Fable 5.2 vs. Astra vs. Carbon

Dimension Fable 5.2 (Anthropic) GPT-6 Astra (OpenAI) Gemini 4 Carbon (Google)
Intelligence Index 59 (#1) 57 ~59 (unreleased)
Standard Price $8/$40 ~$10/$50 ~$15/$75
Cached Price $0.20 ~$1.00 ~$1.50
Speed (vs 5.1) +30% Baseline Unknown
Fast Mode 2.5× at 2× price ~2× at 2× price Not announced
Safety Model Tiered (Fable/Mythos) Opt-in watermarking Fairwind (selective)
Enterprise Features ZDR, Advisor/Orchestrator Dots agents, Wallet Fairwind, Security
Availability General (paid) General Fairwind only
Launch Date Nov 9, 2026 Sep 2026 Oct 28 / Nov ~10

Anthropic's edge: Best price/performance on public benchmarks, immediate general availability, enterprise-ready. OpenAI's edge: Broader ecosystem (Dots, Wallet, Codex), larger user base. Google's edge: Carbon leads on some business-workflow benchmarks (53.9% vs 51.4%), TPU integration.

What This Means for the Market

  1. Inference economics shifting — 20% price cut + 30% speed = ~42% effective cost reduction per task. Opus 5.5 at $4/$20 remains the daily driver; Fable 5.2 is the "power user" tier.
  2. Tiered safety becomes standard — Verification Programs for dangerous capabilities (cyber/bio) is a model other labs may adopt.
  3. Agentic orchestration maturing — 1,000-agent OS build and Fermat proof formalization prove multi-agent coordination at scale.
  4. IPO narrative reinforced — "Pacing, not pausing" with tangible product velocity counters investor concerns about Amodei's slowdown call.

What Comes Next

  • IPO pricing (Nov 18) — Fable 5.2 metrics will feature in roadshow; revenue per token trajectory key metric
  • Gemini 4 Carbon general availability — Will test Anthropic's Intelligence Index lead
  • OpenAI response — Expected GPT-6.1 Sol/Luna pricing adjustments; possible Astra price cut
  • Verification Program expansion — Anthropic may broaden Cyber/Life Sciences access criteria
  • Enterprise adoption — ZDR + Frontier Safeguards + Fast Mode = compelling pitch for regulated industries

Related AIPress Coverage

Sources

  • Anthropic Blog (Nov 9, 2026): "Introducing Claude Fable 5.2 and Claude Mythos 5.2"
  • CNBC (Nov 9, 2026): "Anthropic releases Claude Fable 5.2 and Claude Mythos 5.2 ahead of IPO"
  • TechCrunch (Nov 9, 2026): "Anthropic releases new Mythos and Fable models amid RSI concerns"
  • VentureBeat (Nov 9, 2026): Detailed pricing table, benchmark analysis
  • Fortune (Nov 9, 2026): "Anthropic debuts updated Claude Fable and Mythos models, priced 20% below OpenAI's GPT-6 Astra"
  • The Verge (Nov 9, 2026): "Anthropic launches new Claude Fable and Mythos 5.2 models"
  • The Rundown AI (Nov 9, 2026): "Anthropic says 'pacing, not pausing' with Fable 5.2"
  • DailyAI (Nov 10, 2026): "Claude Fable 5.2 and Mythos 5.2: 20% Cheaper, Up to 30% Faster"
  • Artificial Analysis Intelligence Index (Nov 2026): Leaderboard data

Building something with AI?

DevsIsle designs and ships AI systems, agents and integrations for teams that need it done properly.

Talk to our team →