Skip to contents

Process Document with Claude Opus 4.5 OCR

claude_ocr.Rd

This function processes a document with Claude Opus 4.5 (#1 on OCR Arena leaderboard) and returns structured OCR results in Tensorlake-compatible format. Claude provides exceptional accuracy for complex documents, handwriting, tables, and multi-page PDFs.

Usage

claude_ocr(
  file_path,
  api_key = Sys.getenv("ANTHROPIC_API_KEY"),
  model = "claude-opus-4-6",
  max_tokens = 16000,
  extraction_prompt = NULL,
  output_file = NULL
)

Arguments

file_path

Character string. Path to a local PDF, PNG, JPEG, or image file.

api_key

Character string. Anthropic API key. Default retrieves from environment variable "ANTHROPIC_API_KEY".

model

Character string. Claude model to use. Default is "claude-opus-4-6". Alternative: "claude-sonnet-4-5" for faster/cheaper processing.

max_tokens

Integer. Maximum tokens in response. Default is 16000.

extraction_prompt

Character string. Custom extraction prompt. If NULL, uses default prompt that generates Tensorlake-compatible JSON structure.

output_file

Character string. Optional path to save the JSON response to a file. Default is NULL (no file output).

Value

List. The parsed response from Claude containing:

structured_output

Parsed JSON with pages, tables, and structured data

content

Raw response content from Claude

usage

Token usage information

Note

Claude Opus 4.5 ranks #1 on OCR Arena (ELO: 1696, 71.2% win rate) as of Feb 2026. It excels at:

  • Complex tables and forms

  • Handwritten text

  • Multi-page PDFs

  • Low-quality scans

  • Scientific/technical documents

Pricing varies by model and region. Check Anthropic pricing for current rates.

Author

Nathan C. Layman

Examples

if (FALSE) { # \dontrun{
# Process a PDF with Claude Opus 4.5
result <- claude_ocr("document.pdf")

# Extract pages in Tensorlake format
pages <- claude_extract_pages(result)

# Use faster/cheaper Sonnet model
result <- claude_ocr("document.pdf", model = "claude-sonnet-4.5-20250929")

# Save output to file
result <- claude_ocr("document.pdf", output_file = "ocr_result.json")

# Use with ecoextract
library(ecoextract)
pages <- claude_extract_pages(result)
json_content <- jsonlite::toJSON(pages, auto_unbox = TRUE)
} # }