Claude Haiku 4.5

• Inputs:
  Image, PDF, Text
• Outputs:
  Text

• Maximum input tokens: 200,000
• Maximum output tokens: 64,000

Supported
• Web search previewPreview feature
• Batch predictions
• Prompt caching
• Function calling
• Extended thinking
• Count tokens
Not supported

Supported
• Fixed quota
• Provisioned Throughput
Not supported

• Limitation and specifications: See Vision in Anthropic's documentation

• Limitation and specifications: See PDF support in Anthropic's documentation

• claude-haiku-4.5@20251001
• Launch stage: Generally available
• Release date: October 15, 2025

Model availability

(Includes fixed quota & Provisioned Throughput)

United States
• us-east5
Europe
• europe-west1
Asia pacific
• asia-east1
Global
• global endpoint

ML processing

United States
• Multi-region
Europe
• Multi-region
Asia pacific
• asia-east1

us-east5:

• QPM: 1,500
• Input TPM: 1,500,000 uncached and cache write
• Output TPM: 150,000
• Context length: 200,000

europe-west1:

• QPM: 1,800
• Input TPM: 1,800,000 uncached and cache write
• Output TPM: 180,000
• Context length: 200,000

asia-east1:

• QPM: 2,500
• Input TPM: 2,500,000 uncached and cache write
• Output TPM: 250,000
• Context length: 200,000

global endpoint:

• QPM: 2,500
• Input TPM: 2,500,000 uncached and cache write
• Output TPM: 250,000
• Context length: 200,000