Claude Sonnet 4.6 Thinking

Claude Sonnet 4.6 with extended thinking enabled for tougher coding, planning, and multi‑tool tasks. Ideal for long‑horizon agent workflows and complex problem solving.

Reasoning
Vision
Native PDF input
Tool Calling
Structured Output

Added Feb 17, 2026

Pricing

Auto routing · per 1M tokens

Input: $3.00
Output: $15.00
Cache read: $0.30

Compare provider prices

Specifications

Context window: 1M
Max output: 128K
Avg output (7d): 1K tokens; Longer than 72% of models

Benchmarks

Artificial Analysis

Vectara

Sourced from Artificial Analysis.

Intelligence Index

30.1

Better than 82% of models compared

Coding Index

63.0

Better than 76% of models compared

Agentic Index

31.8

Better than 75% of models compared

Agentic work

AutomationBench-AA

Workflow automation with guardrail penalties

20.1%

Better than 41% of models compared

Harvey LAB-AA

Legal agentic work criterion pass rate

86.0%

Better than 44% of models compared

AA-Briefcase

Agentic knowledge work (Elo)

1062 Elo

Better than 54% of models compared

GDPval-AA v2

Economically valuable tasks (Elo)

1234 Elo

Better than 63% of models compared

Document reasoning

GDP.pdf

Professional PDF reasoning: all-pass rate

15.8%

Better than 59% of models compared

AA-LCR v1.1

Long context reasoning with updated grading

80.0%

Better than 87% of models compared

MLCR-AA

Medical long-context reasoning

24.4%

Better than 77% of models compared

Reasoning

HLE

Humanity's Last Exam

33.6%

Better than 81% of models compared

IFBench

Instruction-following benchmark

56.6%

Better than 67% of models compared

CritPt

Research-level physics reasoning

3.1%

Coding

Terminal-Bench v4.0

Practical coding and terminal tasks

3.0%

Better than 53% of models compared

SciCode

Python programming for scientific computing

50.1%

Better than 53% of models compared

Knowledge

AA-Omniscience Accuracy

Proportion of correctly answered questions

40.9%

AA-Omniscience Hallucination Rate

Rate of incorrect answers among non-correct responses

48.4%

Legacy benchmarks

GPQA Diamond (legacy)

Graduate-level scientific reasoning

87.5%

Better than 84% of models compared

Terminal-Bench Hard (legacy)

Agentic coding and terminal use

53.0%

Better than 96% of models compared

T²-Bench Telecom (legacy)

Conversational AI agents in dual-control scenarios

75.7%

Better than 67% of models compared

AA-LCR (unversioned / legacy)

Long context reasoning evaluation

80.0%

Better than 87% of models compared

GDPval-AA (unversioned / legacy)

Economically valuable tasks

36.0%

Last updated Oct 3, 2026