Claude Sonnet 4.6 vs Claude Opus 4.8

Best cell per row highlighted. Null means undisclosed — never counted as zero.

Anthropic
Claude Sonnet 4.6GA
8.8
AI Panel
Anthropic
Claude Opus 4.8GA
9.2
AI Panel

Identity & lifecycle

ProviderAnthropicAnthropic
Family / tierClaude 4 · SonnetClaude 4 · Opus
StatusGAGA
Released2026-02-162026-05-27
Knowledge cutoff2025-08

Architecture & context

Context window1M1M
Max output tokens64K128K
Input modalitiestext, imagetext, image
Reasoning modeoptionalalways
Open weightsNoNo

Pricing (per Mtok)

Input$3$5
Output$15$25
Cached input$0.3$0.5
Batch input$1.5$2.5
Free tierYesNo

Speed

Output speed (tok/s)7560
Time to first token (s)1.80.5
Latency tierfastmedium

Trust & deployment

Trains on inputsNoNo
LicenseProprietaryProprietary
CloudsBedrock, Vertex AI, Azure AI FoundryBedrock, Vertex AI, Azure AI Foundry
ComplianceSOC2, HIPAA, GDPR, ISO27001SOC2, HIPAA, GDPR, ISO27001

AI Panel scoring

Unified score8.89.2
Decision Maker99
Domain Strategist99
Finance Lead98.5
Domain Practitioner99.5
Power User8.59
Skeptic88
Value score98

Benchmarks

MMLU89.1
MMLU-Pro87.3
GPQA Diamond74.193.6
AIME 202594
MATH-50089
SWE-bench Verified79.688.6
HumanEval98
LiveCodeBench79.7
MMMU83.6
Terminal-Bench74.6
LMArena Elo1460
LMArena Coding Elo1500
Artificial Analysis Index51

Scores link to their sources. Missing cells mean the vendor hasn't disclosed a result — honesty over padding.