Inference
EN
Anthropic

Claude Opus 5.5

Claude Opus 5.5 is a reasoning model from Anthropic. 73 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 100.78.6±2.4Median 51.0
CoverageIndex weight that has results.95%
SpeedOutput tokens per second.76/sMedian 67/s
First tokenSeconds to the first output token.3.4 sMedian 1.3 s
Input / 1MUS dollars per 1M input tokens.$4.00batch $2.00Median $0.60
Output / 1MUS dollars per 1M output tokens.$20.00batch $10.00 Batch price per 1M output tokens.Median $2.55
ContextMaximum tokens in one request.1MMedian 262K
EloLMArena rating and rank.1513±6 #3Median 1422

Details

From OpenRouter
API IDThe model ID on OpenRouter.
anthropic/claude-opus-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.
$0.20
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$4.00$20.001M
AnthropicMaker$4.00$20.001M
DigitalOceanCloud$4.00$20.001M
Google Vertex AICloud$4.00$20.001M
Microsoft AzureCloud$4.00$20.001M
OpenRouterRouter$4.00$20.001M

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.30 counted74.111 / 218
CodingCode writing and repair.17 counted81.82 / 209
ReasoningLogic problems and puzzles.10 counted80.87 / 227
MultimodalTasks with images and text.4 counted74.13 / 84
KnowledgeFacts and expert knowledge.6 counted79.23 / 268
MultilingualTasks in many languages.None counted——
InstructionTasks with strict prompt rules.1 counted59.6—
MathMath problems.5 counted77.74 / 168

Results

73 counted
BenchmarkThe benchmark name.ResultThe score from the source.PlacePlace among results on this benchmark.
OTIS Mock AIME 2024-2025Math100.0%1 / 200
GDPval-AA normalizedAgentic68.3%1 / 133
Artificial Analysis MMMU-ProMultimodal87.7%1 / 113
Artificial Analysis Humanity's Last ExamKnowledge61.4%2 / 222
Terminal-Bench 2.1 · Vals AIAgentic87.6% ±1.71 / 75
LiveBench MathematicsMath97.1%1 / 66
Terminal-Bench 4.0Agentic65.2%1 / 61
Artificial Analysis SciCodeCoding66.9%2 / 114
Artificial Analysis Omniscience AccuracyKnowledge65.5%3 / 164
ProofBench v1.1Math100.0%1 / 54
Critical Physics TasksReasoningAbove fitFar above this model's index.31.7%4 / 214
LiveBench DesignAgentic86.9%1 / 52
LiveBench Long AgenticAgentic85.0%1 / 51
SWE-bench ProCoding89.9%2 / 83
ParseBench: document parsing for AI agents (LlamaIndex leaderboard)Multimodal78.8%1 / 39
Artificial Analysis Long Context ReasoningReasoning84.7%6 / 217
LiveBench CodingCoding89.3%2 / 66
LiveBench Agentic CodingAgentic71.7%2 / 66
Agent Arena task outcomeAgentic20.72 / 54
Humanity's Last Exam without toolsKnowledge64.4%2 / 53
ARC-AGI-1 (semi-private)Reasoning97.5%5 / 126
Agent Arena command recoveryAgentic13.62 / 50
Terminal-Bench 4.1.0Agentic65.8% ±1.61 / 23
Terminal-Bench 4.0.0Agentic64.8% ±1.61 / 23
FrontierMath-Tiers-1-3-v2-PrivateMath91.2% ±1.74 / 90
ARC-AGI-2 (semi-private)Reasoning91.7%6 / 131
Vibe Code Bench v1.1Coding90.3% ±1.56 / 123
LiveBench ReasoningReasoning92.2%4 / 80
MirrorCodeCoding95.6% ±2.62 / 39
Code MigrationCoding66.6% ±4.34 / 77
Furniture AssemblyReasoning83.3% ±4.72 / 37
Humanity's Last Exam with toolsAgentic67.7%2 / 36
Chess PuzzlesReasoning67.0% ±4.75 / 90
Agent Arena steerabilityAgentic18.63 / 54
LiveBench Coding 2.0Coding96.3%3 / 52
FrontierCode 1.1 MainCoding54.4%1 / 17
Hermes Index: agentic capability and cost per task inside Hermes AgentAgentic63.3%1 / 16
PostTrainBench v1.1Coding49.3%1 / 15
FrontierMath-Tier-4-v2-PrivateMath95.0% ±4.15 / 72
FrontierCode 2.0Coding45.5%2 / 27
Mystery Game PuzzlesReasoning71.0% ±4.66 / 80
SimpleQA VerifiedKnowledge72.2% ±1.47 / 91
Vibe Code Bench 1-100Coding30.4% ±4.82 / 26
IOICoding95.1% ±4.94 / 45
OSWorld 2.1 (all 108 tasks)Agentic81.8%1 / 11
CWE-bench v2Agentic67.1%1 / 11
MazeBench: visual spatial reasoning in a 3D open worldAgentic13.0%5 / 48
ProgramBench: Can Language Models Rebuild Programs From Scratch?Coding91.2%2 / 19
HealthBench ProfessionalKnowledge65.6%2 / 18
LiveBench Data AnalysisReasoning80.3%9 / 80
Toolathlon-VerifiedAgentic77.8%3 / 23
LiveBench LanguageKnowledge86.3%9 / 66
FrontierSWE v2Coding62.3%3 / 22
cursorBench40Coding57.8%3 / 21
LiveBench ResearchAgentic85.0%8 / 52
Artificial Analysis AnalystAgentAgentic56.3%4 / 26
Multimodal Knowledge Agent BenchmarkAgentic53.0%3 / 19
EBR-benchReasoning71.4% ±7.15 / 31
DeepSWEAgentic74.2%10 / 59
GPQA diamond · Epoch AIKnowledge90.6% ±1.938 / 217
ScarfBench: Cross-Framework Application Migration in Enterprise JavaCoding70.5%2 / 11
Artificial Analysis Terminal-Bench v4.0Agentic59.6%4 / 22
OfficeQA ProMultimodal67.7%3 / 16
OSWorld 2.1 offline subset (82 tasks)Agentic87.2%2 / 10
Artificial Analysis DeepsecBench-AAAgentic26.9%5 / 21
Artificial Analysis AutomationBenchAgentic69.5%6 / 23
Bug Hunt BenchCoding41.7%8 / 25
LiveBench Browser UseAgentic80.0%17 / 49
ExploitGymAgentic33.3%8 / 23
Artificial Analysis GDP.pdfAgentic26.2%8 / 22
Artificial Analysis CWE-Bench-AAAgentic58.3%7 / 19
Medical Long Context Reasoning (MLCR-AA)Reasoning27.8%9 / 22
CWE-bench v1Agentic67.0%5 / 12
OSWorld 2.0Agentic48.7%11 / 25
Terminal-Bench-Science 0.1Agentic58.7%5 / 11
LiveBench Instruction FollowingInstruction65.7%39 / 66
AutomationBenchAgentic40.0%20 / 32
ExtractBench: schema-guided enterprise document extraction (LlamaIndex leaderboard)Multimodal86.1%17 / 26
Artificial Analysis Cyber IndexAgentic28.7%14 / 19
Artificial Analysis CyberGym-E2E-AAAgentic0.8%15 / 19
Artificial Analysis ITBench-AAAgentic38.2%20 / 24
MedScribeHealthcare91.4% ±1.91 / 109
ARC-AGI-2 (public eval)Reasoning97.9%3 / 122
EMBFinance75.9% ±2.42 / 75
ProgramBenchCoding18.5% ±2.82 / 65
SAGEEducation91.3% ±1.41 / 29
ARC-AGI-1 (public eval)Reasoning98.3%8 / 116
Tax Agent BenchFinance44.5% ±3.66 / 82
Time Horizon Index: KSPOther91.3%1 / 13
Legal Research BenchLegal50.5% ±3.57 / 90
Public Benefits Bench v1.1Public benefits70.6% ±1.24 / 50
Terminal-Bench-ScienceScience47.1% ±6.04 / 44
MysteryMechanismScience49.5% ±3.43 / 30
SRE BenchOther33.6% ±2.94 / 36
Finance Agent (v2)Finance58.6% ±0.210 / 79
BioMysteryBenchScience79.3% ±1.04 / 30
FrontierMath-ErdosMath2.9% ±2.12 / 15
MedCodeHealthcare49.8% ±2.319 / 106
CUA-benchOther14.0%2 / 9
Vals Web Search IndexOther53.7% ±2.03 / 12
SpireClimb (Slay the Spire 1, Ascension 20 Heart)Other5.0% ±2.85 / 9
HLAB-Vals-1.1Legal9.2% ±2.422 / 35
SpireClimb (Slay the Spire 2, Ascension 10)Other6.7% ±2.96 / 9
SpireClimbOther5.8% ±2.06 / 9
CyberBench v1.1Other55.4% ±5.140 / 48
ArXivMath August 2026 with toolsMath96.9%—
BenchCAD Vision2Code voxel IoU with toolsMultimodal96.2%—
Global MMLUMultilingual94.3%—
Multi-task Indic Language Understanding BenchmarkMultilingual93.1%—
Artificial Analysis Harvey LAB-AA v1.0Agentic91.2%—
Legal Agent Benchmark mean criterion-pass rate — Harvey held-out setAgentic91.2%—
ArXivMath August 2026 without toolsMath91.2%—
BioMysteryBench Human SolvableKnowledge89.3%—
Chartography with image and code toolsMultimodal89.0%—
Data Research and Analysis with Complex OperationsAgentic87.4%—
Anthropic de novo protein-binder design evaluationKnowledge82.6%—
Toolathlon Verified Pass@3Agentic82.4%—
Anthropic CVE RediscoveryAgentic81.3%—
OfficeQAMultimodal78.9%—
HealthBench Professional raw scoreKnowledge77.1%—
Hermes Bench: 150 agent tasks graded on the files and state the agent leaves behindAgentic76.7%—
Molecular Biology Protocols TroubleshootingKnowledge73.7%—
Protein domain identificationKnowledge73.6%—
BenchCAD Vision2Code voxel IoU without toolsMultimodal73.0%—
WANDR (Wide ANd Deep Research)Agentic72.3%—
Toolathlon Verified Pass cubedAgentic72.2%—
LatchBio SpatialBench VerifiedKnowledge72.0%—
Anthropic biomedical-image-analysis evaluationMultimodal71.4%—
Artificial Analysis HealthBench Professional raw scoreKnowledge71.4%—
Benchling Molecular Biology Protocols UnderstandingKnowledge69.0%—
PhysicianBenchKnowledge68.4%—
HealthBench raw scoreKnowledge68.1%—
GraphWalks BFS 256K–1MReasoning66.8%—
BenchCAD 2.1 Core, with toolsMultimodal65.3%—
Chartography without toolsMultimodal64.4%—
Artificial Analysis HealthBench raw scoreKnowledge64.3%—
FrontierCode 1.1 ExtendedCoding63.6%—
Anthropic medicinal-chemistry evaluationKnowledge63.5%—
AA-Briefcase v1.1, rubric pass rateAgentic61.2%—
LatchBio SingleCellBenchKnowledge61.2%—
HealthBench length-adjusted scoreKnowledge60.6%—
Anthropic Protein Design evaluationKnowledge60.2%—
Artificial Analysis HealthBench Professional length-adjusted scoreKnowledge59.9%—
Artificial Analysis HealthBench length-adjusted scoreKnowledge57.0%—
Anthropic protein-design library-ranking taskKnowledge56.0%—
Anthropic STEM Professor BenchReasoning53.7%—
ExtractBench Pro: long-tail document extraction challenges (LlamaIndex)Multimodal51.0%—
BioMysteryBench Human DifficultKnowledge50.0%—
Protein mutation attributionKnowledge49.5%—
BenchCAD 2.1 Core, without toolsMultimodal49.0%—
RoboDojo SimAgentic39.4%—
Axiom Bio morphology-to-molecule matchingKnowledge34.0%—
Toolathlon Verified average assistant turnsAgentic26.9%—
Remote Labor Index (RLI)Agentic21.3%—
Legal Agent Benchmark all-pass rate — Harvey held-out setAgentic8.3%—
Harvey LAB-AA v1.1Agentic4.2%—
STEMBench120: six-domain scientific research progressReasoning2.6%—

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
Index81.5 → 78.6
23 Oct 202612 Nov 2026
Output / 1M$20.00 → $20.00
23 Oct 202612 Nov 2026

Questions

How much does Claude Opus 5.5 cost?
$4.00 per 1M input tokens, $20.00 per 1M output tokens.
How good is Claude Opus 5.5?
Its index is 78.6 out of 100. That is place 3 of 283 ranked models.
What is Claude Opus 5.5 best at?
Coding: 81.8 out of 100, place 2 of 209.
How fast is Claude Opus 5.5?
76 output tokens per second, 3.4 s to the first token.
What is the context window of Claude Opus 5.5?
1M tokens.
Is Claude Opus 5.5 open source?
No. Its weights are not published.
When was Claude Opus 5.5 released?
September 2026.

Badge

Claude Opus 5.5 on Inference360: place 3 of 283, index 78.6
HTML
<a href="https://inference360.ai/models/claude-opus-5-5"><img src="https://inference360.ai/badges/claude-opus-5-5.svg" alt="Claude Opus 5.5 on Inference360: place 3 of 283, index 78.6" height="20"></a>
Markdown
[![Claude Opus 5.5 on Inference360: place 3 of 283, index 78.6](https://inference360.ai/badges/claude-opus-5-5.svg)](https://inference360.ai/models/claude-opus-5-5)

Compare with

Gemini 4 Argon78.5GPT-6 Astra78.0GPT-6.1 Sol78.0Muse Wonder Full77.3Claude Sonnet 5.577.2

Lower price

GPT-6.1 Terra77.1 · $5.00/1MGemini 4 Carbon82.7 · $10.00/1MGemini 4 Argon78.5 · $10.00/1M

More from Anthropic

All models
Claude Sonnet 5.577.2Claude Haiku 5.566.3Claude Fable 5.176.2Claude Mythos 5.1UnrankedClaude Fable 5.285.6