Claude Haiku 4.5

[reasoning]

Anthropic[text]

Sonnet-4-level coding performance at 1/3 cost and 2x speed

Input

per 1M tokens

Output

per 1M tokens

Context

200K tokens

Max Output

64K tokens

Released

2025-10-15

Features

[optional-reasoning][computer-use][agent-based-tasks][coding][safest-model]

Technical Specifications

Benchmarks

AIME 202580.7%

GPQA Diamond73.0%

Humanity's Last Exam9.7%

MMMLU83.0%

SWE-bench Verified73.3%

Tau2-bench83.2%

The Lab

location: San Francisco, California, USAfounded: 2021backed by: Amazon, Google; public benefit corporation

Founded in 2021 by former OpenAI researchers and headquartered in San Francisco, Anthropic is a public benefit corporation backed by Amazon and Google. The lab pursues safety-first AI research using Constitutional AI to align models with human values. Its flagship Claude family emphasises long-context understanding, nuanced instruction-following, and steerable behaviour.

more about Anthropic →

Model ID: claude-haiku-4-5