Inference
EN
Anthropic

Claude Sonnet 5.5

Claude Sonnet 5.5 is a reasoning model from Anthropic. 59 benchmarks count toward its index, in 7 categories.

  • Closed weights
  • Reasoning
  • Input: text · image · file
  • Released 2026-09
RankedEnough results for an index.
IndexAll categories, out of 100.77.2±2.6Median 51.0
CoverageIndex weight that has results.95%
SpeedOutput tokens per second.107/sMedian 67/s
First tokenSeconds to the first output token.1.6 sMedian 1.3 s
Input / 1MUS dollars per 1M input tokens.$2.00batch $1.00Median $0.60
Output / 1MUS dollars per 1M output tokens.$10.00batch $5.00 Batch price per 1M output tokens.Median $2.55
ContextMaximum tokens in one request.1MMedian 262K
EloLMArena rating and rank.1469±6 #34Median 1422

Details

From OpenRouter
API IDThe model ID on OpenRouter.
anthropic/claude-sonnet-5.5
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.
$0.10
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$2.00$10.001M
AnthropicMaker$2.00$10.001M
DigitalOceanCloud$2.00$10.001M
Google Vertex AICloud$2.00$10.001M
Microsoft AzureCloud$2.00$10.001M
OpenRouterRouter$2.00$10.001M

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.20 counted77.55 / 218
CodingCode writing and repair.14 counted80.83 / 209
ReasoningLogic problems and puzzles.10 counted77.810 / 227
MultimodalTasks with images and text.3 counted74.42 / 84
KnowledgeFacts and expert knowledge.6 counted71.314 / 268
MultilingualTasks in many languages.None counted——
InstructionTasks with strict prompt rules.1 counted41.5—
MathMath problems.5 counted76.110 / 168

Results

59 counted
BenchmarkThe benchmark name.ResultThe score from the source.PlacePlace among results on this benchmark.
OTIS Mock AIME 2024-2025Math100.0%1 / 200
Code MigrationCoding69.8% ±4.31 / 77
LiveBench CodingCoding91.4%1 / 66
Vibe Code Bench v1.1Coding92.4% ±1.32 / 123
GPQA diamond · Epoch AIKnowledge95.6% ±1.44 / 217
ProofBench v1.1Math100.0%1 / 54
Agent Arena command recoveryAgentic15.01 / 50
MirrorCodeCoding97.8% ±1.51 / 39
Artificial Analysis Humanity's Last ExamKnowledge55.0%6 / 222
GDPval-AA normalizedAgentic66.9%4 / 133
Critical Physics TasksReasoningAbove fitFar above this model's index.31.4%7 / 214
SWE-bench ProCoding81.3%3 / 83
LiveBench DesignAgentic85.0%2 / 52
Terminal-Bench 4.0Agentic64.1% ±1.03 / 61
Artificial Analysis SciCodeCoding61.0%6 / 114
HealthBench ProfessionalKnowledge69.2%1 / 18
Agent Arena task outcomeAgentic19.53 / 54
LiveBench Long AgenticAgentic78.8%3 / 51
ExtractBench: schema-guided enterprise document extraction (LlamaIndex leaderboard)Multimodal94.7%2 / 26
Artificial Analysis AnalystAgentAgentic57.5%2 / 26
Terminal-Bench 2.1 · Vals AIAgentic83.1% ±1.76 / 75
Bug Hunt BenchCoding51.3%2 / 25
Terminal-Bench 4.1.0Agentic63.9% ±1.52 / 23
Terminal-Bench 4.0.0Agentic61.8% ±1.52 / 23
ARC-AGI-1 (semi-private)Reasoning96.5%11 / 126
LiveBench ReasoningReasoning91.6%7 / 80
Mystery Game PuzzlesReasoning65.0% ±4.87 / 80
Medical Long Context Reasoning (MLCR-AA)Reasoning41.7%2 / 22
Humanity's Last Exam without toolsKnowledge56.9%5 / 53
Artificial Analysis Omniscience AccuracyKnowledge51.5%17 / 164
Artificial Analysis Long Context ReasoningReasoning82.7%23 / 217
FrontierMath-Tiers-1-3-v2-PrivateMath88.8% ±1.910 / 90
Humanity's Last Exam with toolsAgentic64.5%4 / 36
Chess PuzzlesReasoning59.0% ±4.910 / 90
LiveBench Coding 2.0Coding94.7%6 / 52
LiveBench MathematicsMath96.1%8 / 66
Agent Arena steerabilityAgentic14.57 / 54
ARC-AGI-2 (semi-private)Reasoning83.9%17 / 131
Toolathlon-VerifiedAgentic77.8%3 / 23
Artificial Analysis Terminal-Bench v4.0Agentic63.6%3 / 22
FrontierCode 2.0Coding41.0%4 / 27
FrontierMath-Tier-4-v2-PrivateMath80.5% ±6.311 / 72
FrontierSWE v2Coding61.9%4 / 22
Furniture AssemblyReasoning75.0% ±5.57 / 37
cursorBench40Coding55.5%4 / 21
LiveBench ResearchAgentic83.6%10 / 52
IOICoding83.1% ±3.79 / 45
Multimodal Knowledge Agent BenchmarkAgentic52.0%4 / 19
Artificial Analysis AutomationBenchAgentic71.8%5 / 23
LiveBench Browser UseAgentic84.0%11 / 49
MazeBench: visual spatial reasoning in a 3D open worldAgentic8.0%11 / 48
Hermes Index: agentic capability and cost per task inside Hermes AgentAgentic53.1%4 / 16
ARC-AGI-3 (semi-private)Reasoning24.3%7 / 26
OSWorld 2.1 (all 108 tasks)Agentic80.1%3 / 11
OSWorld 2.1 offline subset (82 tasks)Agentic83.9%3 / 10
ProgramBench: Can Language Models Rebuild Programs From Scratch?Coding79.7%6 / 19
ParseBench: document parsing for AI agents (LlamaIndex leaderboard)Multimodal69.2%13 / 39
DeepSWEAgentic71.0%21 / 59
Terminal-Bench-Science 0.1Agentic59.9%4 / 11
LiveBench Agentic CodingAgentic56.3%27 / 66
SimpleQA VerifiedKnowledge46.5% ±1.639 / 91
Artificial Analysis GDP.pdfAgentic25.8%10 / 22
FrontierCode 1.1 MainCoding46.2%8 / 17
OfficeQA ProMultimodal65.6%8 / 16
LiveBench LanguageKnowledge78.0%38 / 66
LiveBench Instruction FollowingInstruction56.8%60 / 66
LiveBench Data AnalysisReasoning59.5%73 / 80
MedScribeHealthcare91.1% ±2.03 / 109
EMBFinance75.7% ±2.43 / 75
ProgramBenchCoding6.5% ±1.74 / 65
BioMysteryBenchScience81.1% ±0.62 / 30
Terminal-Bench-ScienceScience47.1% ±6.04 / 44
MedCodeHealthcare52.9% ±2.112 / 106
ARC-AGI-1 (public eval)Reasoning97.8%14 / 116
Tax Agent BenchFinance42.0% ±3.610 / 82
ARC-AGI-2 (public eval)Reasoning90.1%16 / 122
MysteryMechanismScience49.1% ±3.44 / 30
Legal Research BenchLegal48.1% ±3.512 / 90
FrontierMath-ErdosMath2.9% ±2.12 / 15
Finance Agent (v2)Finance58.1% ±0.712 / 79
SRE BenchOther30.2% ±2.88 / 36
Public Benefits Bench v1.1Public benefits67.2% ±1.215 / 50
SAGEEducation88.0% ±1.610 / 29
HLAB-Vals-1.1Legal10.4% ±2.620 / 35
CyberBench v1.1Other59.6% ±5.237 / 48
SpireClimb (Slay the Spire 1, Ascension 20 Heart)Other0.0%7 / 9
SpireClimb (Slay the Spire 2, Ascension 10)Other2.7% ±1.98 / 9
SpireClimbOther1.3% ±0.98 / 9
BenchCAD Vision2Code voxel IoU with toolsMultimodal96.3%—
ArXivMath August 2026 with toolsMath95.2%—
Legal Agent Benchmark mean criterion-pass rate — Harvey held-out setAgentic93.1%—
Artificial Analysis Harvey LAB-AA v1.0Agentic93.1%—
Global MMLUMultilingual92.1%—
Multi-task Indic Language Understanding BenchmarkMultilingual91.6%—
Chartography with image and code toolsMultimodal90.2%—
BioMysteryBench Human SolvableKnowledge89.2%—
Data Research and Analysis with Complex OperationsAgentic87.0%—
ArXivMath August 2026 without toolsMath86.8%—
Toolathlon Verified Pass@3Agentic85.2%—
Anthropic de novo protein-binder design evaluationKnowledge82.3%—
HealthBench Professional raw scoreKnowledge77.1%—
OfficeQAMultimodal76.9%—
BenchCAD Vision2Code voxel IoU without toolsMultimodal74.7%—
Hermes Bench: 150 agent tasks graded on the files and state the agent leaves behindAgentic73.5%—
LatchBio SpatialBench VerifiedKnowledge72.5%—
Anthropic biomedical-image-analysis evaluationMultimodal72.2%—
Artificial Analysis HealthBench Professional raw scoreKnowledge70.8%—
HealthBench raw scoreKnowledge69.4%—
Toolathlon Verified Pass cubedAgentic68.5%—
Molecular Biology Protocols TroubleshootingKnowledge67.3%—
Benchling Molecular Biology Protocols UnderstandingKnowledge66.6%—
Artificial Analysis HealthBench raw scoreKnowledge65.9%—
HealthBench length-adjusted scoreKnowledge65.4%—
Anthropic medicinal-chemistry evaluationKnowledge65.3%—
PhysicianBenchKnowledge63.2%—
Artificial Analysis HealthBench Professional length-adjusted scoreKnowledge63.0%—
Artificial Analysis HealthBench length-adjusted scoreKnowledge62.1%—
Chartography without toolsMultimodal61.6%—
FrontierCode 1.1 ExtendedCoding59.1%—
LatchBio SingleCellBenchKnowledge59.1%—
Anthropic protein-design library-ranking taskKnowledge54.8%—
Anthropic Protein Design evaluationKnowledge51.0%—
AutomationBench, Zapier private evaluation set release 1.0.6Agentic44.7%—
BioMysteryBench Human DifficultKnowledge44.7%—
ExtractBench Pro: long-tail document extraction challenges (LlamaIndex)Multimodal32.0%—
Toolathlon Verified average assistant turnsAgentic31.6%—
Axiom Bio morphology-to-molecule matchingKnowledge25.0%—
Legal Agent Benchmark all-pass rate — Harvey held-out setAgentic10.0%—

Sources

BenchLMCC BY-NC 4.0OpenRouterTerms restrict reuseVals AINo licence statedEpoch AICC BYARC Prize FoundationTerms restrict reuseLiveBenchNo licence statedTerminal-BenchNo licence statedLMArenaCC BY 4.0models.devMIT

History

5 releases
Index78.4 → 77.2
23 Oct 202612 Nov 2026
Output / 1M$10.00 → $10.00
23 Oct 202612 Nov 2026

Questions

How much does Claude Sonnet 5.5 cost?
$2.00 per 1M input tokens, $10.00 per 1M output tokens.
How good is Claude Sonnet 5.5?
Its index is 77.2 out of 100. That is place 8 of 283 ranked models.
What is Claude Sonnet 5.5 best at?
Coding: 80.8 out of 100, place 3 of 209.
How fast is Claude Sonnet 5.5?
107 output tokens per second, 1.6 s to the first token.
What is the context window of Claude Sonnet 5.5?
1M tokens.
Is Claude Sonnet 5.5 open source?
No. Its weights are not published.
When was Claude Sonnet 5.5 released?
September 2026.

Badge

Claude Sonnet 5.5 on Inference360: place 8 of 283, index 77.2
HTML
<a href="https://inference360.ai/models/claude-sonnet-5-5"><img src="https://inference360.ai/badges/claude-sonnet-5-5.svg" alt="Claude Sonnet 5.5 on Inference360: place 8 of 283, index 77.2" height="20"></a>
Markdown
[![Claude Sonnet 5.5 on Inference360: place 8 of 283, index 77.2](https://inference360.ai/badges/claude-sonnet-5-5.svg)](https://inference360.ai/models/claude-sonnet-5-5)

Compare with

Muse Wonder Full77.3GPT-6.1 Terra77.1GPT-6.1 Sol78.0GPT-6 Astra78.0Claude Fable 5.176.2

Lower price

GPT-6.1 Terra77.1 · $5.00/1M

More from Anthropic

All models
Claude Opus 5.578.6Claude Haiku 5.566.3Claude Fable 5.176.2Claude Mythos 5.1UnrankedClaude Fable 5.285.6