Back to the index
Anthropic · Claude
No. 04

Claude Sonnet 4.6

The production workhorse — best balance of speed and intelligence.

BalancedClosed weightsText, Image, Pdf
Field notes
Context window
1Mtokens
Max output
64Ktokens
Input price
$3/ 1M tokens
Output price
$15/ 1M tokens
Cached input
$0.30/ 1M tokens
Modalities
Text, Image, Pdf
Knowledge cutoff
Nov 2025
Released
Feb 2026
Approx. speed
Overview

What this model is

Claude Sonnet 4.6 offers the best balance of speed and intelligence in the Claude lineup and is the default production workhorse for most agentic and coding workloads. It handles the large majority of real tasks reliably at a fraction of Opus pricing, escalating to Opus or Fable only for the hardest jobs. Its 1M context and strong tool use make it a dependable everyday driver.

Strengths

  • Excellent speed-to-intelligence balance
  • Reliable everyday agentic coding
  • 1M context at a moderate price
  • Strong tool use and instruction following

Trade-offs

  • Below Opus/Fable on the hardest long-horizon jobs
  • 64K max output cap
  • No audio/video input

Best for

  • Everyday production coding agents
  • High-volume developer workflows
  • Chat and assistant backends
  • RAG and document analysis

Not ideal for

  • The most demanding autonomous coding runs
  • Tasks needing very long single outputs
Capabilities

Capability profile

Normalized 0–100 scores, comparable across the whole catalog.

Reasoning89Coding89Math86Writing90Knowledge90Speed78Agentic88Vision84Multilingual89Long Context92
Reasoning
89
Coding
89
Math
86
Writing
90
Knowledge
90
Speed
78
Agentic
88
Vision
84
Multilingual
89
Long Context
92
Benchmarks

How it scores

Public benchmark results, with independent third-party results where available. Bars normalize percentages to 100 and Elo ratings to a 1500 ceiling.

Independent index
Artificial Analysis Intelligence Index
Composite of ~9–10 independent evals · Artificial Analysis
47/ 100
BenchmarkResult
Artificial Analysis Intelligence Indexindependent
GeneralArtificial Analysis
47%
SWE-bench Verifiedindependent
CodingSWE-bench
79.6%
τ-benchindependent
Agenticτ-bench
87.5%
LiveBenchindependent
GeneralLiveBench
75.5%
GPQA Diamondindependent
ReasoningIndependent aggregators
89.9%
DeepSWEindependent
CodingDatacurve
32%
LMArena Eloindependent
GeneralLMArena
1467Elo
Alternatives

Comparable models

Models in a roughly similar class — worth weighing against this one.