Tech Meridian← ALL MODELS
PROMY MERIDIAN

ANTHROPIC · MODEL RELEASE TRACKER

Claude Fable 5.2

Anthropic's Claude Fable 5.2 is a named model/version referenced in an Artificial Analysis evaluation. The evaluation compared it with OpenAI's GPT-6.2 Astra and Sol and found Astra and Sol scored lower on structure and transparency than Claude Fable 5.2.

CURRENT SNAPSHOT4/5 DIMENSIONS WITH DATA

The dimensions that change the decision.

PRICING
Estimated cost per task (reported in comparison)$26.54 per task (reported alongside the benchmark comparison; Kimi K3.1 Pro was reported at $3.68 per task)
CONTEXT WINDOW

Not established from the available sources.

MODALITIES
Modalities demonstratedVisual/image generation combined with code generation/execution (painting via generated Python code)
Modality: language model that outputs code and does not rely on external image modelsThe article describes Fable 5.2 as a language model that creates images without using an external image model or pre-rendered/traced source images, instead emitting Python programs that drive a simulated paint engine.
BENCHMARKS
Inference speedAnthropic says Fable 5.2 returns responses about 30% faster across effort levels than prior versions; an even faster inference tier is available at double the price.
Benchmark ranking (Artificial Analysis AI Index v5)Reported to rank first with 65 points on the Artificial Analysis AI Index v5 (five points ahead of Gemini 4 Carbon and 15 ahead of Fable 5.1); 62 points at lower compute tiers per the same index reporting.
Inference speedReported to return responses about 30% faster across effort levels compared with Fable 5.1; an optional faster mode can increase token output speed (reported up to 2.5×) at higher cost and limited platform availability.
MathArena rankingPlaced 2nd on the MathArena math benchmark (behind Gemini 4 Carbon).
RoboDojo Sim benchmark scoreOn Anthropic's internal RoboDojo Sim benchmark, Fable 5.2 scored 64.9%, compared with 29.8% for OpenAI's GPT-6 Astra.
Artificial Analysis Intelligence Index score65 (Claude Fable 5.2 scored 65 on the Artificial Analysis Intelligence Index)
Score on the Artificial Analysis Intelligence Index65 (Claude Fable 5.2 scored 65 on the Artificial Analysis Intelligence Index, compared with 46 for Kimi K3.1 Pro)
Included in Artificial Analysis benchmarkClaude Fable 5.2 was included in an Artificial Analysis evaluation that compared models on structure, transparency, and other personality indexes.
Higher structure/transparency score vs GPT-6.2 Astra/SolIn the Artificial Analysis evaluation, Claude Fable 5.2 scored higher on structure and transparency than OpenAI's GPT-6.2 Astra and Sol.
AVAILABILITY
AvailabilityDescribed in the article in the context of being among publicly available models.

RELEASE TIMELINE

Published, source-backed release events only.

4 SOURCES · IMPORTANCE 8.0

Anthropic releases Claude Fable 5.2; Claude Mythos 5.2 available to vetted orgs

Anthropic announced two new models, Claude Fable 5.2 and Claude Mythos 5.2, claiming improvements in coding, reasoning, and vision and faster output than the 5.1 versions. Anthropic says Fable 5.2 is broadly available today while Mythos 5.2 access is limited to vetted organizations, and that prerelease testing helped solve complex engineering and scientific tasks (including work on black hole physics and formalizing Fermat’s Last Theorem).

→

VERIFIABLE FACTS

Every value stays attached to a source and date.

RELEASE · Release and availabilityDEVELOPER CLAIM

Announced and made available on November 9, 2026; launched on Amazon Bedrock and the Claude Platform on AWS and presented as broadly available to users.

CAPABILITIES · Positioning for technical workDEVELOPER CLAIM

Anthropic describes Fable 5.2 as its most capable generally available model for technical work, raising the bar in coding and the sciences.

CAPABILITIES · Long-running and self-verifying operationDEVELOPER CLAIM

Handles ambiguous, multi-step work; verifies its own output; can run unattended for hours to days; interprets prompts, iterates, catches its own errors, and optimizes as it goes.

CAPABILITIES · Designed for demanding technical work (coding, math, science)DEVELOPER CLAIM

Positioned as Anthropic's most capable generally available model for technical work: strong performance on coding, reasoning, and scientific tasks; takes on ambiguous, multi-step work, verifies its own output, and can run unattended for hours to days.

CAPABILITIES · Reported research achievementsDEVELOPER CLAIM

During prerelease testing, reported to have helped solve two open problems in theoretical physics and to have formalized Fermat's Last Theorem (claims reported by Anthropic and media coverage).

BENCHMARKS · Inference speedDEVELOPER CLAIM

Reported to return responses about 30% faster across effort levels compared with Fable 5.1; an optional faster mode can increase token output speed (reported up to 2.5×) at higher cost and limited platform availability.

MODALITIES · Modality: language model that outputs code and does not rely on external image modelsDEVELOPER CLAIM

The article describes Fable 5.2 as a language model that creates images without using an external image model or pre-rendered/traced source images, instead emitting Python programs that drive a simulated paint engine.

WHAT CHANGED

Stored passport versions, without reconstructed history.

Passport updated28 facts
BENCHMARKS · Higher structure/transparency score vs GPT-6.2 Astra/Sol+ In the Artificial Analysis evaluation, Claude Fable 5.2 scored higher on structure and transparency than OpenAI's GPT-6.2 Astra and Sol.
BENCHMARKS · Included in Artificial Analysis benchmark+ Claude Fable 5.2 was included in an Artificial Analysis evaluation that compared models on structure, transparency, and other personality indexes.
Passport updated26 facts
BENCHMARKS · Score on the Artificial Analysis Intelligence Index+ 65 (Claude Fable 5.2 scored 65 on the Artificial Analysis Intelligence Index, compared with 46 for Kimi K3.1 Pro)
PRICING · Estimated cost per task (reported in comparison)+ $26.54 per task (reported alongside the benchmark comparison; Kimi K3.1 Pro was reported at $3.68 per task)
PRICING · Cost per task in comparison on the AAAI Index− $26.54 per task (Claude Fable 5.2 cost per task in the cited comparison)
Passport updated25 facts
BENCHMARKS · Artificial Analysis Intelligence Index score+ 65 (Claude Fable 5.2 scored 65 on the Artificial Analysis Intelligence Index)
PRICING · Cost per task in comparison on the AAAI Index+ $26.54 per task (Claude Fable 5.2 cost per task in the cited comparison)
MODALITIES · Modalities demonstrated+ Visual/image generation combined with code generation/execution (painting via generated Python code)
CAPABILITIES · Generated photorealistic artwork by writing and executing Python code+ Claude Fable 5.2 painted photorealistic artwork stroke by stroke using self-written Python code
PRICING · Relative cost in tests− With the reasoning settings used in testing, cost was roughly 40 times that of GPT-6.1 Sol.
MODALITIES · Modalities− Reported improvements in coding, reasoning, and vision capabilities (indicating multimodal capabilities including vision in addition to text).
Passport updated23 facts
CAPABILITIES · Examples and scale of paintings produced+ In demonstrations the model produced a parking-garage scene using roughly 700,000 simulated brushstrokes and a bird painting from 111 coordinate sequences; paintings are driven by Python programs of about 1,000–2,000 lines.
MODALITIES · Modality: language model that outputs code and does not rely on external image models+ The article describes Fable 5.2 as a language model that creates images without using an external image model or pre-rendered/traced source images, instead emitting Python programs that drive a simulated paint engine.
SAFETY · Behavior to mitigate imitation concerns+ Following criticism of an earlier sibling model for imitating existing art, Fable 5.2 was described as developing its own artist personas with fictional backstories and consistent styles to address such concerns.
CAPABILITIES · Image generation via simulated painting (code-driven)+ Fable 5.2 generates images by producing Python code that places brushstrokes on a virtual canvas, simulating physical painting properties such as bristle texture and paint application.
LIMITATIONS · Real-world robot test: speed and fine motor control limitations+ Although Fable 5.2 scored highly on RoboDojo Sim, it reportedly fell behind a competitor in speed and fine motor control during a real-world test with a physical robot arm.
LIMITATIONS · Resource intensity: many brushstrokes and long programs+ The approach requires several thousand brushstrokes per painting (examples include hundreds of thousands) and Python programs of roughly 1,000–2,000 lines to drive the simulated paint engine.
BENCHMARKS · RoboDojo Sim benchmark score+ On Anthropic's internal RoboDojo Sim benchmark, Fable 5.2 scored 64.9%, compared with 29.8% for OpenAI's GPT-6 Astra.
Passport updated16 facts
BENCHMARKS · MathArena ranking+ Placed 2nd on the MathArena math benchmark (behind Gemini 4 Carbon).
CAPABILITIES · Math capability+ Described as potentially one of the most capable publicly available models for math tasks.
AVAILABILITY · Availability+ Described in the article in the context of being among publicly available models.
PRICING · Relative cost in tests+ With the reasoning settings used in testing, cost was roughly 40 times that of GPT-6.1 Sol.
PRICING · Pricing (token-based)− Reported pricing: $8 per million input tokens and $40 per million output tokens (a reduction from prior $10/$50 levels); one report notes the input-token price was lowered from $10 to $8.
Passport updated13 facts
CAPABILITIES · Agentic coding and long-running agents+ Supports agentic coding across large, multi-repository codebases and long-running agents for coordinating many subagents on multi-day work.
BENCHMARKS · Benchmark ranking (Artificial Analysis AI Index v5)+ Reported to rank first with 65 points on the Artificial Analysis AI Index v5 (five points ahead of Gemini 4 Carbon and 15 ahead of Fable 5.1); 62 points at lower compute tiers per the same index reporting.
BENCHMARKS · Inference speed+ Reported to return responses about 30% faster across effort levels compared with Fable 5.1; an optional faster mode can increase token output speed (reported up to 2.5×) at higher cost and limited platform availability.
RELEASE · Release and availabilityAnnounced and made available November 9, 2026; available on Amazon Bedrock and Claude Platform on AWS.→Announced and made available on November 9, 2026; launched on Amazon Bedrock and the Claude Platform on AWS and presented as broadly available to users.
CAPABILITIES · Reported research achievements+ During prerelease testing, reported to have helped solve two open problems in theoretical physics and to have formalized Fermat's Last Theorem (claims reported by Anthropic and media coverage).
CAPABILITIES · Designed for demanding technical work (coding, math, science)+ Positioned as Anthropic's most capable generally available model for technical work: strong performance on coding, reasoning, and scientific tasks; takes on ambiguous, multi-step work, verifies its own output, and can run unattended for hours to days.
PRICING · Pricing (token-based)Anthropic lowered Fable's price from $10 to $8 per million input tokens for Fable 5.2.→Reported pricing: $8 per million input tokens and $40 per million output tokens (a reduction from prior $10/$50 levels); one report notes the input-token price was lowered from $10 to $8.
MODALITIES · Modalities+ Reported improvements in coding, reasoning, and vision capabilities (indicating multimodal capabilities including vision in addition to text).
MODALITIES · Modality: vision improvements− Reports state Fable 5.2 shows improvements in vision in addition to coding and reasoning.
Passport created8 facts