OCR (Optical Character Recognition)

LLMBase does not have a separate OCR endpoint. Text extraction works through the standard chat completions API: send an image as part of a user message to a vision-capable model, and ask it to return the text it reads.

Try it without an account

The OCR demo lets you upload a receipt, invoice, screenshot, or PDF and see the extracted result in your browser, with no account and no API key required. It is rate-limited and intended for evaluation, not production traffic.

Extracting text via the API

Send the image as an image_url content part alongside a text instruction. This example uses qwen/qwen3-vl-30b-a3b-instruct, the vision-capable model used by the public OCR demo:

import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://api.llmbase.ai/v1",
  apiKey: process.env.LLMBASE_API_KEY,
});

const response = await client.chat.completions.create({
  model: "qwen/qwen3-vl-30b-a3b-instruct",
  temperature: 0,
  messages: [
    {
      role: "user",
      content: [
        {
          type: "text",
          text: "Extract all readable text from this image. Return only the extracted text, preserving line breaks.",
        },
        {
          type: "image_url",
          image_url: { url: "data:image/png;base64,<base64-encoded-image>" },
        },
      ],
    },
  ],
});

console.log(response.choices[0].message.content);

image_url.url accepts either a data URI (as above) or a publicly reachable https:// URL. See Chat completions for the full request and response reference.

Structured extraction

For invoices, receipts, and forms, request JSON output instead of a text blob with response_format, and describe the fields you want in the prompt:

const response = await client.chat.completions.create({
  model: "qwen/qwen3-vl-30b-a3b-instruct",
  temperature: 0,
  response_format: { type: "json_object" },
  messages: [
    {
      role: "user",
      content: [
        {
          type: "text",
          text: 'Extract this invoice as JSON with keys: vendor, invoiceNumber, date, total, lineItems (array of {description, amount}). Use an empty string for missing fields. Do not invent values.',
        },
        { type: "image_url", image_url: { url: "data:image/png;base64,<base64-encoded-image>" } },
      ],
    },
  ],
});

See Structured outputs for JSON mode and JSON Schema details.

Choosing a vision-capable model

Not every model accepts image input. Check input_modalities and supported_input_image_media_types before sending a request:

curl "https://api.llmbase.ai/v1/models?metadata=true" \
  -H "Authorization: Bearer $LLMBASE_API_KEY"

If a model does not advertise image support, LLMBase returns a 400 error instead of silently ignoring the image. See Model discovery for the full metadata reference.

Supported inputs

• Images — JPEG, PNG, and WebP are broadly supported; confirm exact formats against the selected model’s supported_input_image_media_types.
• PDFs — the chat completions API accepts image content parts, not PDF files directly. Render each PDF page you want to process to an image first, then send it the same way as a photo or screenshot.
• Handwriting — vision models can attempt handwritten text, but accuracy is lower than for printed text. Treat handwritten results as a starting point and verify against the source.

Authentication

Use your LLMBase API key in the Authorization header, the same as any other Inference API request:

Authorization: Bearer <LLMBASE_API_KEY>

See Authentication for how to create and manage API keys.