Skip to main content

​
Overview

OpenAI provides vision-capable models that can process documents and images. Zerox supports all OpenAI GPT-4 vision models for OCR and data extraction tasks.

​
Credentials

To use OpenAI models, you need to provide your API key:
​
credentials.apiKey
string
required
Your OpenAI API key. Can be obtained from the OpenAI API dashboard.

​
Environment Variable

You can store your API key as an environment variable: 
export OPENAI_API_KEY="sk-..."

​
Supported Models

The following OpenAI models are available through Zerox:
ModelModel IDDescription
GPT-4.1gpt-4.1Latest GPT-4.1 model with vision capabilities
GPT-4.1 Minigpt-4.1-miniSmaller, faster GPT-4.1 model
GPT-4ogpt-4oOptimized GPT-4 model with vision
GPT-4o Minigpt-4o-miniSmaller, cost-effective GPT-4o model

​
Configuration

​
Basic Example

import { zerox } from "zerox";
import { ModelOptions, ModelProvider } from "zerox/node-zerox/dist/types";

const result = await zerox({
  filePath: "path/to/document.pdf",
  modelProvider: ModelProvider.OPENAI,
  model: ModelOptions.OPENAI_GPT_4O,
  credentials: {
    apiKey: process.env.OPENAI_API_KEY,
  },
});

​
With Environment Variable

import { zerox } from "zerox";

const result = await zerox({
  filePath: "path/to/document.pdf",
  credentials: {
    apiKey: process.env.OPENAI_API_KEY,
  },
});
When using environment variables, you can omit the modelProvider parameter. OpenAI is the default provider.

​
LLM Parameters

OpenAI models support the following optional parameters:
​
llmParams.temperature
number
default:"1"
Controls randomness in the output. Values range from 0 to 2. Lower values make output more focused and deterministic.
​
llmParams.maxTokens
number
default:"4096"
Maximum number of tokens to generate in the completion.
​
llmParams.topP
number
default:"1"
Nucleus sampling parameter. An alternative to temperature sampling. Values range from 0 to 1.
​
llmParams.frequencyPenalty
number
default:"0"
Number between -2.0 and 2.0. Positive values penalize tokens based on their frequency in the text so far.
​
llmParams.presencePenalty
number
default:"0"
Number between -2.0 and 2.0. Positive values penalize tokens based on whether they appear in the text so far.
​
llmParams.logprobs
boolean
default:"false"
Whether to return log probabilities of the output tokens. Useful for confidence scoring.

​
Example with Parameters

import { zerox } from "zerox";
import { ModelOptions, ModelProvider } from "zerox/node-zerox/dist/types";

const result = await zerox({
  filePath: "path/to/document.pdf",
  modelProvider: ModelProvider.OPENAI,
  model: ModelOptions.OPENAI_GPT_4O_MINI,
  credentials: {
    apiKey: process.env.OPENAI_API_KEY,
  },
  llmParams: {
    temperature: 0.3,
    maxTokens: 4096,
    topP: 0.9,
    logprobs: true,
  },
});

​
Data Extraction

OpenAI models support structured data extraction using JSON schemas: 
import { zerox } from "zerox";
import { ModelOptions, ModelProvider } from "zerox/node-zerox/dist/types";

const schema = {
  type: "object",
  properties: {
    invoice_number: { type: "string" },
    date: { type: "string" },
    total: { type: "number" },
    items: {
      type: "array",
      items: {
        type: "object",
        properties: {
          description: { type: "string" },
          quantity: { type: "number" },
          price: { type: "number" },
        },
      },
    },
  },
};

const result = await zerox({
  filePath: "invoice.pdf",
  modelProvider: ModelProvider.OPENAI,
  model: ModelOptions.OPENAI_GPT_4O,
  credentials: {
    apiKey: process.env.OPENAI_API_KEY,
  },
  extractOnly: true,
  schema: schema,
});

console.log(result.extracted);

​
Error Handling

import { zerox } from "zerox";
import { ErrorMode, ModelProvider } from "zerox/node-zerox/dist/types";

try {
  const result = await zerox({
    filePath: "document.pdf",
    modelProvider: ModelProvider.OPENAI,
    credentials: {
      apiKey: process.env.OPENAI_API_KEY,
    },
    errorMode: ErrorMode.THROW,
  });
} catch (error) {
  console.error("OCR failed:", error);
}
Ensure your OpenAI API key has sufficient credits and appropriate permissions for vision API access. Vision models consume tokens based on image resolution and document complexity.

​
Best Practices

  • Use gpt-4o-mini for cost-effective processing of simple documents
  • Use gpt-4o or gpt-4.1 for complex layouts, tables, and detailed extraction
  • Set temperature: 0 for deterministic output when consistency is critical
  • Enable logprobs: true to access confidence scores for extraction quality assessment
  • Monitor token usage through the returned inputTokens and outputTokens fields

Build docs developers (and LLMs) love

Get started for freeTalk to us