Refractr · House #32,056

What it is

Calls itself “document extraction API”.

• Calls itself: document extraction API

“Refractr is a document extraction API that converts unstructured data into validated JSON.”

Where it puts itself

Claims “Fastest”.

• Superlative claim: Fastest

“A custom-trained extraction model, not a general-purpose LLM wrapped in a prompt.”

“Fastest structured JSON extraction from documents”

What it costs

A free tier up to a usage limit. Prices quoted: €1 = 200 credits, €5.

• Free plan: free up to a limit
• Price: €1 = 200 credits · €5

“€1 = 200 credits · top up from €5”

“…top up from €5”

Who it sells to

Sells to developers.

• Sells to: developers

“Refractr is a document extraction API that converts unstructured data into validated JSON.”

How it runs

A API. Hosted in the cloud. Offers an API. AI is the core of the product.

• Kind: api
• Runs where: cloud / hosted
• API: yes
• AI role: core to the product

“Refractr is a document extraction API that converts unstructured data into validated JSON.”

“Refractr runs a custom, optimized model hosted by us, not an OpenAI or Anthropic passthrough.”

What it works with

Names AWS Textract, DeepSeek V4.1 Flash and 5 more on its pages.

• Names: AWS Textract · DeepSeek V4.1 Flash · GLiNER2 (base) · +4

“OCR speed: Refractr 0.77s, AWS Textract 2.9s, Google Document AI 3.8s, Azure Document Intelligence 4.3s, GPT-4o reading the image 16.9s, GPT-4o + separate OCR 33.0s.”

“Accuracy: Gemini 3.8 Flash 85.1, DeepSeek V4.1 Flash 82.2, Refractr 82.0, GPT-5.4 mini 77.3, Llama 3.1 8B 66.1, NuExtract 2.0 (2B) 65.4, GLiNER2 (base) 44.7.”

What it offers as proof

In beta. Trying it asks for an email. Publishes docs.

• Stage: beta
• Instant signup: yes
• Trial asks for: an email address

“Alpha release”

“Sign up free”