Claude Haiku 4.5 Thinking

Claude Haiku 4.5 with extended thinking enabled for deeper planning, more deliberate coding, and multi-step reasoning without moving up to Sonnet pricing.

Reasoning
Vision
Native PDF input
Tool Calling
Structured Output

Added Oct 15, 2025

Pricing

Auto routing · per 1M tokens

Input: $1.00
Output: $5.00
Cache read: $0.10

Compare provider prices

Specifications

Context window: 200K
Max output: 64K
Avg output (7d): 657 tokens; Longer than 53% of models

Benchmarks

Artificial Analysis

Vectara

Sourced from Artificial Analysis.

Intelligence Index

16.9

Better than 60% of models compared

Coding Index

43.9

Better than 51% of models compared

Agentic Index

8.0

Better than 42% of models compared

Agentic work

AutomationBench-AA

Workflow automation with guardrail penalties

3.2%

Better than 25% of models compared

Harvey LAB-AA

Legal agentic work criterion pass rate

61.1%

Better than 11% of models compared

AA-Briefcase

Agentic knowledge work (Elo)

616 Elo

Better than 27% of models compared

GDPval-AA v2

Economically valuable tasks (Elo)

719 Elo

Better than 34% of models compared

Document reasoning

GDP.pdf

Professional PDF reasoning: all-pass rate

3.8%

Better than 20% of models compared

AA-LCR v1.1

Long context reasoning with updated grading

74.3%

Better than 72% of models compared

MLCR-AA

Medical long-context reasoning

6.1%

Better than 19% of models compared

Reasoning

HLE

Humanity's Last Exam

10.4%

Better than 51% of models compared

IFBench

Instruction-following benchmark

54.3%

Better than 65% 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

42.2%

Better than 32% of models compared

LiveCodeBench

Contamination-free coding benchmark

61.5%

Better than 67% of models compared

Math

AIME 2025

American Invitational Mathematics Examination 2025

83.7%

Better than 80% of models compared

Knowledge

MMLU-Pro

Professional and academic subject knowledge

76.0%

Better than 52% of models compared

Legacy benchmarks

GPQA Diamond (legacy)

Graduate-level scientific reasoning

67.2%

Better than 44% of models compared

Terminal-Bench Hard (legacy)

Agentic coding and terminal use

27.3%

Better than 69% of models compared

T²-Bench Telecom (legacy)

Conversational AI agents in dual-control scenarios

54.7%

Better than 55% of models compared

AA-LCR (unversioned / legacy)

Long context reasoning evaluation