Inference
EN
Anthropic

Claude Fable 5.2

Claude Fable 5.2 is a reasoning model from Anthropic in the Claude Fable family. 37 benchmarks count toward its index, in 6 categories.

  • New
  • SmartestHighest index.
  • Closed weights
  • Reasoning
  • Input: text · image · file
  • Released 2026-11
RankedEnough results for an index.
IndexAll categories, out of 100.85.6±3.7Median 51.0
CoverageIndex weight that has results.90%
SpeedOutput tokens per second.64/sMedian 67/s
First tokenSeconds to the first output token.4.9 sMedian 1.3 s
Input / 1MUS dollars per 1M input tokens.$8.00batch $4.00Median $0.60
Output / 1MUS dollars per 1M output tokens.$40.00batch $20.00 Batch price per 1M output tokens.Median $2.55
ContextMaximum tokens in one request.1MMedian 262K
EloLMArena rating and rank.—Median 1422

Details

From OpenRouter
API IDThe model ID on OpenRouter.
anthropic/claude-fable-5.2
Max outputMaximum output tokens per request.
128K
Knowledge cutoffLast date of training data.
—
ToolsTool calls on OpenRouter.
Yes
JSON outputOutput follows a given JSON schema.
Yes
Cache read / 1MUS dollars per 1M cached tokens.
$0.20
WeightsLink to the published weights.
—
RetiresDate OpenRouter removes the model.
—

Where to run

APIs that serve this model.
6 providers
ProviderTypeMaker, cloud, host or router.InputOutputContext
Amazon BedrockCloud$8.00$40.001M
AnthropicMaker$8.00$40.001M
DigitalOceanCloud$8.00$40.001M
Google Vertex AICloud$8.00$40.001M
Microsoft AzureCloud$8.00$40.001M
OpenRouterRouter$8.00$40.001M

US dollars per 1M tokens, list price. From models.dev (opens in a new tab) (MIT).

Capabilities

Index per category, out of 100.
Out of 100
AgenticMulti-step tasks with tools.10 counted82.22 / 218
CodingCode writing and repair.8 counted86.21 / 209
ReasoningLogic problems and puzzles.10 counted86.51 / 227
MultimodalTasks with images and text.1 counted82.0—
KnowledgeFacts and expert knowledge.4 counted87.11 / 268
MultilingualTasks in many languages.None counted——
InstructionTasks with strict prompt rules.None counted——
MathMath problems.4 counted79.31 / 168

Results

37 counted
BenchmarkThe benchmark name.ResultThe score from the source.PlacePlace among results on this benchmark.
Artificial Analysis Humanity's Last ExamKnowledge74.3%1 / 222
Critical Physics TasksReasoning35.4%1 / 214
Artificial Analysis Omniscience AccuracyKnowledge79.1%1 / 164
Vibe Code Bench v1.1Coding93.8% ±1.21 / 123
Artificial Analysis SciCodeCoding67.7%1 / 114
SimpleQA VerifiedKnowledge94.0% ±0.81 / 91
FrontierMath-Tiers-1-3-v2-PrivateMath95.1% ±1.31 / 90
Chess PuzzlesReasoning79.0% ±4.11 / 90
SWE-bench ProCoding93.0%1 / 83
Mystery Game PuzzlesReasoning90.0% ±3.01 / 80
Humanity's Last ExamKnowledge74.2%1 / 77
FrontierMath-Tier-4-v2-PrivateMath100.0%1 / 72
Terminal-Bench 4.0Agentic65.2% ±2.31 / 61
ProofBench v1.1Math100.0%1 / 54
Agent Arena steerabilityAgentic33.01 / 54
Agent Arena task outcomeAgentic22.01 / 54
Humanity's Last Exam without toolsKnowledge73.2%1 / 53
ARC-AGI-2 (semi-private)Reasoning92.5%3 / 131
Furniture AssemblyReasoning90.0% ±3.91 / 37
EBR-benchReasoning90.5% ±1.51 / 31
FrontierCode 2.0Coding49.1%1 / 27
ARC-AGI-3 (semi-private)Reasoning65.7%1 / 26
ARC-AGI-1 (semi-private)Reasoning97.5%5 / 126
Bug Hunt BenchCoding58.0%1 / 25
FrontierSWE v2Coding69.9%1 / 22
Medical Long Context Reasoning (MLCR-AA)Reasoning43.9%1 / 22
Multimodal Knowledge Agent BenchmarkAgentic56.6%1 / 19
Artificial Analysis Long Context ReasoningReasoning83.3%13 / 217
OfficeQA ProMultimodal72.9%1 / 16
GPQA diamond · Epoch AIKnowledge93.9% ±1.614 / 217
OTIS Mock AIME 2024-2025Math99.7% ±0.314 / 200
cursorBench40Coding60.0%2 / 21
OSWorld 2.1 offline subset (82 tasks)Agentic89.3%1 / 10
Artificial Analysis AutomationBenchAgentic76.4%3 / 23
DeepSWEAgentic74.3%9 / 59
Terminal-Bench-Science 0.1Agentic74.6%2 / 11
Artificial Analysis Terminal-Bench v4.0Agentic59.6%4 / 22
ProgramBench: Can Language Models Rebuild Programs From Scratch?Coding85.6%4 / 19
AutomationBenchAgentic51.4%11 / 32
Artificial Analysis GDP.pdfAgentic24.2%12 / 22
ARC-AGI-2 (public eval)Reasoning98.8%1 / 122
ProgramBenchCoding19.0% ±2.81 / 65
ARC-AGI-1 (public eval)Reasoning99.0%2 / 116
BioMysteryBenchScience83.7% ±0.71 / 30
Terminal-Bench-ScienceScience55.7% ±6.02 / 44
MysteryMechanismScience51.8% ±3.42 / 30
Legal Research BenchLegal46.2% ±3.519 / 90
SpireClimb (Slay the Spire 1, Ascension 20 Heart)Other21.7% ±5.42 / 9
Tax Agent BenchFinance32.1% ±3.423 / 82
SpireClimbOther19.6% ±3.54 / 9
SpireClimb (Slay the Spire 2, Ascension 10)Other17.6% ±4.54 / 9
HLAB-Vals-1.1Legal5.0% ±1.727 / 35
Global MMLUMultilingual96.0%—
Multi-task Indic Language Understanding BenchmarkMultilingual94.1%—
Chartography with image and code toolsMultimodal91.6%—
Data Research and Analysis with Complex OperationsAgentic87.7%—
BenchCAD 2.1 Core, with toolsMultimodal83.5%—
OfficeQAMultimodal83.4%—
WANDR (Wide ANd Deep Research)Agentic82.1%—
BenchCAD 2.1 Core, without toolsMultimodal75.4%—
Chartography without toolsMultimodal73.4%—
Artificial Analysis HealthBench Professional raw scoreKnowledge71.1%—
Artificial Analysis HealthBench Professional length-adjusted scoreKnowledge67.7%—
AA-Briefcase v1.1, rubric pass rateAgentic66.1%—
RoboDojo SimAgentic64.9%—
PhysicianBenchKnowledge64.4%—
Anthropic STEM Professor BenchReasoning61.8%—
Artificial Analysis HealthBench length-adjusted scoreKnowledge59.6%—
Artificial Analysis HealthBench raw scoreKnowledge58.8%—
Remote Labor Index (RLI)Agentic41.3%—
Harvey LAB-AA v1.1Agentic7.2%—

Sources

BenchLMCC BY-NC 4.0OpenRouterTerms restrict reuseVals AINo licence statedEpoch AICC BYARC Prize FoundationTerms restrict reuseLMArenaCC BY 4.0models.devMIT

History

5 releases
Index85.4 → 85.6
11 Nov 202612 Nov 2026
Output / 1M$40.00 → $40.00
11 Nov 202612 Nov 2026

Questions

How much does Claude Fable 5.2 cost?
$8.00 per 1M input tokens, $40.00 per 1M output tokens.
How good is Claude Fable 5.2?
Its index is 85.6 out of 100. That is place 1 of 283 ranked models.
What is Claude Fable 5.2 best at?
Knowledge: 87.1 out of 100, place 1 of 268.
How fast is Claude Fable 5.2?
64 output tokens per second, 4.9 s to the first token.
What is the context window of Claude Fable 5.2?
1M tokens.
Is Claude Fable 5.2 open source?
No. Its weights are not published.
When was Claude Fable 5.2 released?
November 2026.

Badge

Claude Fable 5.2 on Inference360: place 1 of 283, index 85.6
HTML
<a href="https://inference360.ai/models/claude-fable-5-2"><img src="https://inference360.ai/badges/claude-fable-5-2.svg" alt="Claude Fable 5.2 on Inference360: place 1 of 283, index 85.6" height="20"></a>
Markdown
[![Claude Fable 5.2 on Inference360: place 1 of 283, index 85.6](https://inference360.ai/badges/claude-fable-5-2.svg)](https://inference360.ai/models/claude-fable-5-2)

Compare with

Gemini 4 Carbon82.7Claude Opus 5.578.6Gemini 4 Argon78.5GPT-6 Astra78.0GPT-6.1 Sol78.0

Lower price

Gemini 4 Carbon82.7 · $10.00/1M

More from Anthropic

All models
Claude Mythos 5.2UnrankedClaude Haiku 5.566.3Claude Sonnet 5.577.2Claude Opus 5.578.6Claude Fable 5.176.2