Claude Sonnet 4.5 Thinking

Adds extended, step‑by‑step reasoning for tougher coding, planning, and multi‑tool tasks. Ideal for long‑horizon agent workflows, complex problem solving, and scenarios that benefit from explicit thinking traces.

Reasoning
Vision
Native PDF input
Tool Calling
Structured Output

Added Sep 29, 2025

Pricing

Auto routing · per 1M tokens

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

Compare provider prices

Specifications

Context window: 200K
Max output: 64K
Avg output (7d): 1K tokens; Longer than 72% of models

Benchmarks

Artificial Analysis

Vectara

Design Arena

Sourced from Artificial Analysis.

Intelligence Index

20.7

Better than 69% of models compared

Coding Index

52.1

Better than 60% of models compared

Agentic Index

15.8

Better than 50% of models compared

Agentic work

AutomationBench-AA

Workflow automation with guardrail penalties

14.1%

Better than 37% of models compared

AA-Briefcase

Agentic knowledge work (Elo)

711 Elo

Better than 29% of models compared

GDPval-AA v2

Economically valuable tasks (Elo)

911 Elo

Better than 40% of models compared

Document reasoning

GDP.pdf

Professional PDF reasoning: all-pass rate

5.2%

Better than 25% of models compared

AA-LCR v1.1

Long context reasoning with updated grading

72.3%

Better than 67% of models compared

Reasoning

HLE

Humanity's Last Exam

17.8%

Better than 64% of models compared

IFBench

Instruction-following benchmark

57.3%

Better than 68% of models compared

Coding

Terminal-Bench v4.0

Practical coding and terminal tasks

0.0%

Better than 16% of models compared

SciCode

Python programming for scientific computing

45.7%

Better than 40% of models compared

LiveCodeBench

Contamination-free coding benchmark

71.4%

Better than 82% of models compared

Math

AIME 2025

American Invitational Mathematics Examination 2025

88.0%

Better than 86% of models compared

Knowledge

MMLU-Pro

Professional and academic subject knowledge

87.5%

Better than 98% of models compared

Legacy benchmarks

GPQA Diamond (legacy)

Graduate-level scientific reasoning

83.4%

Better than 73% of models compared

Terminal-Bench Hard (legacy)

Agentic coding and terminal use

35.6%

Better than 82% of models compared

T²-Bench Telecom (legacy)

Conversational AI agents in dual-control scenarios

78.1%

Better than 68% of models compared

AA-LCR (unversioned / legacy)

Long context reasoning evaluation