Extract more exact field values than Claude Sonnet 5.5, 88% cheaper

Gemma 4 31B reads checks, invoices, statements and forms into typed JSON that fits your schema. SIE is the open-source inference engine that serves this vision model.

Get started Run in your cloud

SIE Gemma 4 31B

Specs match the note

Delivery note line item 1 description: rated power output 662W continuous; 220-240v input, 24v DC output; power factor correction 0.98, efficiency rating 91%; recessed mounting design; compact steel enclosure, weight 1.4kg; IP65 protection class. — Delivery note, item 1

items[0].productSpecsJson

{
  "power": {
    "watts": 662, "efficiency": 91,
    "powerFactor": 0.98, "inputVoltage": "220-240v"
  },
  "ratings": { "protection": "ip65" },
  "physical": {
    "weight": 1.4, "material": "steel",
    "mountingType": "recessed"
  }
}

Delivery note line item 1: rated power output 662W continuous; 220-240v input, 24v DC output; power factor correction 0.98, efficiency rating 91%. — Delivery note, item 1

items[0].productSpecsJson.power

{
  "watts": 662,
  "efficiency": 91,
  "powerFactor": 0.98,
  "inputVoltage": "220-240v"
}

Wrong bank details become payment errors

GPT-6.1 Sol (max reasoning)

Wrong check number

Printed routing · account · check numbers

Bank check, scanned sideways

Returned fields

{ "checkNumber": 6335 }

A wrong check number breaks reconciliation against the issued payment.

Claude Haiku 5.5 (high effort)

Wrong routing and account numbers

Printed routing · account · check numbers

Bank check, scanned sideways

Returned fields

{
  "routingNumber": 449263305,
  "accountNumber": 51242541359
}

Wrong bank numbers can send a payment to the wrong destination.

arXiv:2607.21617 Amazon study: AI document readers turn unusual words into familiar ones. GPT-5.4 mini did it to 1 in 4.
Do VLMs Read or Rewrite? On Transcription Faithfulness in Vision-Language Models

Lee, Ak, Mohta, Xu and Dimitriadis, Amazon

“they often tend to rewrite it into a more plausible form”
29 May 2026
ACL Anthology 2026.findings-acl.1237 Findings of ACL 2026 ACL 2026 study: every vision model tested reported numbers it could not see
Perceptual Hallucination in Vision–Language Models: Definition, Analysis and Verification

Hwang, Heo, Lee and Jung

“perceptual hallucination occurs across all models”
11 June 2026

98% of GPT-6.1 Sol's accuracy, 85% cheaper

Run this eval

Exact field values

546 documents ↑

GPT-6.1 Sol

GPT-6.1 Sol 96.1%

SIE Gemma 4 31B

SIE Gemma 4 31B 94.3%

GPT-6 Luna (high reasoning)

GPT-6 Luna (high reasoning) 92.3%

Claude Sonnet 5.5

Claude Sonnet 5.5 92.2%

Claude Haiku 5.5

Claude Haiku 5.5 88.2%

Mistral OCR 4.1

Mistral OCR 4.1 83.4%

Cost per 1,000 documents

USD ↓

GPT-6 Luna (high reasoning)

$1.12 GPT-6 Luna (high reasoning)

Claude Haiku 5.5

$1.19 Claude Haiku 5.5

SIE Gemma 4 31B

$2.12 SIE Gemma 4 31B

Mistral OCR 4.1

$5.00 Mistral OCR 4.1

GPT-6.1 Sol

GPT-6.1 Sol $14.63

Claude Sonnet 5.5

Claude Sonnet 5.5 $18.68

Evaluation

Tested: OmniAI OCR Benchmark: 546 business documents in 17 formats
Scored: Answer-key fields returned exactly; failed or cut-off replies score zero
Prices: Recorded tokens; Serverless and provider list prices

Send a page and your schema, get typed fields back

from pathlib import Path
import base64
import json
from sie_sdk import SIEClient
client = SIEClient(api_key="API keysk-sie-…")
schema = json.loads("""
schema{
  "type": "object",
  "properties": {
    "items": {"anyOf": [{"type": "array", "items": {"type": "object", "properties": {···"brand": ···, "quantity": ···, "reference": ···, "productSpecsJson": {···"watts": ···, "efficiency": ···}, {"type": "null"}]},
    "header": {"type": "object", "properties": {···"date": ···}
  },
  "required": ["items", "header"],
  "additionalProperties": false
}
""")
instruction = "instruction···Copy values as they appear on the document.···"
prompt = instruction
image = base64.b64encode(Path(fileJPGdelivery-note.jpgbrowse).read_bytes()).decode("ascii")
# Messages: the prompt as the system message, the image as the user message. Copy includes them.
result = client.chat_completions(
    "modelgoogle/gemma-4-31B-it:h100-96k-hires-out8k-no-spec",
    messages,
    max_completion_tokens=8192,
    extra_body={"temperature":0,"presence_penalty":0},
    # Strict JSON-schema response_format. Copy includes it.
)
# Copy also reads and parses the JSON reply.

import { readFile } from 'node:fs/promises';
import { SIEClient } from '@superlinked/sie-sdk';

const client = new SIEClient(undefined, { apiKey: 'sk-sie-…' });

const schema = {
  "type": "object",
  "properties": {
    "items": {"anyOf": [{"type": "array", "items": {"type": "object", "properties": {···"brand": ···, "quantity": ···, "reference": ···, "productSpecsJson": {···"watts": ···, "efficiency": ···}, {"type": "null"}]},
    "header": {"type": "object", "properties": {···"date": ···}
  },
  "required": ["items", "header"],
  "additionalProperties": false
};
const instruction = "···Copy values as they appear on the document.···";
const prompt = instruction;
const image = await readFile("delivery-note.jpg");
const result = await client.chatCompletions({
  model: 'google/gemma-4-31B-it:h100-96k-hires-out8k-no-spec',
  // Messages: the prompt as the system message, the image as the user message. Copy includes them.
  max_completion_tokens: 8192,
  temperature: 0,
  presence_penalty: 0,
  // Strict JSON-schema response_format. Copy includes it.
});
// Copy also reads and parses the JSON reply.

document_image=$(base64 < 'delivery-note.jpg' | tr -d '\n')
curl https://api.superlinked.com/v1/chat/completions \
  -H "Authorization: Bearer sk-sie-…" \
  -H "Content-Type: application/json" \
  -d "{\"model\":\"google/gemma-4-31B-it:h100-96k-hires-out8k-no-spec\",\"messages\":[{\"role\":\"system\",\"content\":\"···Copy values as they appear on the document.···\"},{\"role\":\"user\",\"content\":[{\"type\":\"image_url\",\"image_url\":{\"url\":\"data:image/jpeg;base64,$document_image\"}}]}],\"max_completion_tokens\":8192,\"temperature\":0,\"presence_penalty\":0,\"response_format\":{\"type\":\"json_schema\",\"json_schema\":{\"name\":\"fields\",\"strict\":true,\"schema\":{\"type\":\"object\",\"properties\":{\"items\":{\"anyOf\":[{\"type\":\"array\",\"items\":{\"type\":\"object\",\"properties\":{\"sn\":{\"anyOf\":[{\"type\":\"number\"},{\"type\":\"null\"}]},\"type\":{\"anyOf\":[{\"type\":\"string\",\"enum\":[\"LED Luminaire\",\"Power Supply\",\"Control System\",\"Mounting Bracket\",\"LED Driver\",\"Cable Assembly\",\"Junction Box\",\"Sensor Module\"]},{\"type\":\"null\"}]},\"unit\":{\"anyOf\":[{\"type\":\"string\",\"enum\":[\"pcs\",\"set\",\"unit\"]},{\"type\":\"null\"}]},\"brand\":{\"anyOf\":[{\"type\":\"string\",\"enum\":[\"Philips\",\"Osram\",\"Schneider Electric\",\"Siemens\",\"ABB\",\"Lutron\",\"Legrand\",\"Mean Well\"]},{\"type\":\"null\"}]},\"quantity\":{\"anyOf\":[{\"type\":\"number\"},{\"type\":\"null\"}]},\"reference\":{\"anyOf\":[{\"type\":\"string\"},{\"type\":\"null\"}]},\"productSpecsJson\":{\"type\":\"object\",\"properties\":{\"power\":{\"type\":\"object\",\"properties\":{\"watts\":{\"anyOf\":[{\"type\":\"number\"},{\"type\":\"null\"}]},\"efficiency\":{\"anyOf\":[{\"description\":\"Efficiency in percentage, 0-100\",\"type\":\"number\"},{\"type\":\"null\"}]},\"powerFactor\":{\"anyOf\":[{\"type\":\"number\"},{\"type\":\"null\"}]},\"inputVoltage\":{\"anyOf\":[{\"type\":\"string\",\"enum\":[\"100-240v\",\"100-277v\",\"220-240v\"]},{\"type\":\"null\"}]}},\"required\":[\"watts\",\"efficiency\",\"powerFactor\",\"inputVoltage\"],\"additionalProperties\":false},\"ratings\":{\"type\":\"object\",\"properties\":{\"protection\":{\"anyOf\":[{\"type\":\"string\",\"enum\":[\"ip65\",\"ip66\",\"ip67\"]},{\"type\":\"null\"}]}},\"required\":[\"protection\"],\"additionalProperties\":false},\"physical\":{\"type\":\"object\",\"properties\":{\"weight\":{\"anyOf\":[{\"type\":\"number\"},{\"type\":\"null\"}]},\"material\":{\"anyOf\":[{\"type\":\"string\",\"enum\":[\"aluminum\",\"steel\",\"plastic\"]},{\"type\":\"null\"}]},\"mountingType\":{\"anyOf\":[{\"type\":\"string\",\"enum\":[\"surface\",\"recessed\",\"track\"]},{\"type\":\"null\"}]}},\"required\":[\"weight\",\"material\",\"mountingType\"],\"additionalProperties\":false}},\"required\":[\"power\",\"ratings\",\"physical\"],\"additionalProperties\":false}},\"required\":[\"sn\",\"type\",\"unit\",\"brand\",\"quantity\",\"reference\",\"productSpecsJson\"],\"additionalProperties\":false}},{\"type\":\"null\"}]},\"header\":{\"type\":\"object\",\"properties\":{\"to\":{\"type\":\"object\",\"properties\":{\"tel\":{\"anyOf\":[{\"type\":\"string\"},{\"type\":\"null\"}]},\"email\":{\"anyOf\":[{\"type\":\"string\"},{\"type\":\"null\"}]},\"location\":{\"anyOf\":[{\"type\":\"string\"},{\"type\":\"null\"}]},\"attention\":{\"anyOf\":[{\"type\":\"string\"},{\"type\":\"null\"}]},\"reference\":{\"anyOf\":[{\"type\":\"string\"},{\"type\":\"null\"}]},\"vatNumber\":{\"anyOf\":[{\"type\":\"string\"},{\"type\":\"null\"}]},\"companyName\":{\"anyOf\":[{\"type\":\"string\"},{\"type\":\"null\"}]}},\"required\":[\"tel\",\"email\",\"location\",\"attention\",\"reference\",\"vatNumber\",\"companyName\"],\"additionalProperties\":false},\"date\":{\"anyOf\":[{\"type\":\"string\"},{\"type\":\"null\"}]},\"from\":{\"type\":\"object\",\"properties\":{\"tel\":{\"anyOf\":[{\"type\":\"string\"},{\"type\":\"null\"}]},\"email\":{\"anyOf\":[{\"type\":\"string\"},{\"type\":\"null\"}]},\"location\":{\"anyOf\":[{\"type\":\"string\"},{\"type\":\"null\"}]},\"reference\":{\"anyOf\":[{\"type\":\"string\"},{\"type\":\"null\"}]},\"vatNumber\":{\"anyOf\":[{\"type\":\"string\"},{\"type\":\"null\"}]},\"companyName\":{\"anyOf\":[{\"type\":\"string\"},{\"type\":\"null\"}]}},\"required\":[\"tel\",\"email\",\"location\",\"reference\",\"vatNumber\",\"companyName\"],\"additionalProperties\":false},\"projectName\":{\"anyOf\":[{\"type\":\"string\"},{\"type\":\"null\"}]}},\"required\":[\"to\",\"date\",\"from\",\"projectName\"],\"additionalProperties\":false}},\"required\":[\"items\",\"header\"],\"additionalProperties\":false}}}}"

Build the "Doc field extraction" capability into my app using the Superlinked Inference Engine (SIE).

Context
- SIE is an OpenAI-style inference API. Python SDK: `from sie_sdk import SIEClient`; TypeScript: `@superlinked/sie-sdk`.
- Base URL: https://api.superlinked.com (or my regional endpoint). Auth: Bearer key from env `SIE_API_KEY` (never hard-code it).
- Model: google/gemma-4-31B-it:h100-96k-hires-out8k-no-spec (OpenAI-compatible endpoint: /v1/chat/completions). Keep the model id configurable.

Task
- Input: an uploaded document plus the JSON schema for its fields.
- Behaviour: parse the document, then return fields matching the JSON schema
- Send one POST /v1/chat/completions request (SDK: chat_completions / chatCompletions) with a system message and a user message. The system message is "Extract data from the document image into the JSON schema. Copy values as they appear on the document. Use null for a field the document does not contain. Return only the JSON.\n\nJSON schema:\n{\n  \"type\": \"object\",\n  \"required\": [\n    \"header\",\n    \"items\"\n  ],\n  \"properties\": {\n    \"items\": {\n      \"type\": \"array\",\n      \"items\": {\n        \"type\": \"object\",\n        \"required\": [\n          \"sn\",\n          \"type\",\n          \"brand\",\n          \"reference\",\n          \"productSpecsJson\",\n          \"unit\",\n          \"quantity\"\n        ],\n        \"properties\": {\n          \"sn\": {\n            \"type\": \"number\"\n          },\n          \"type\": {\n            \"enum\": [\n              \"LED Luminaire\",\n              \"Power Supply\",\n              \"Control System\",\n              \"Mounting Bracket\",\n              \"LED Driver\",\n              \"Cable Assembly\",\n              \"Junction Box\",\n              \"Sensor Module\"\n            ],\n            \"type\": \"string\"\n          },\n          \"unit\": {\n            \"enum\": [\n              \"pcs\",\n              \"set\",\n              \"unit\"\n            ],\n            \"type\": \"string\"\n          },\n          \"brand\": {\n            \"enum\": [\n              \"Philips\",\n              \"Osram\",\n              \"Schneider Electric\",\n              \"Siemens\",\n              \"ABB\",\n              \"Lutron\",\n              \"Legrand\",\n              \"Mean Well\"\n            ],\n            \"type\": \"string\"\n          },\n          \"quantity\": {\n            \"type\": \"number\"\n          },\n          \"reference\": {\n            \"type\": \"string\"\n          },\n          \"productSpecsJson\": {\n            \"type\": \"object\",\n            \"required\": [\n              \"power\",\n              \"physical\",\n              \"ratings\"\n            ],\n            \"properties\": {\n              \"power\": {\n                \"type\": \"object\",\n                \"required\": [\n                  \"watts\",\n                  \"inputVoltage\"\n                ],\n                \"properties\": {\n                  \"watts\": {\n                    \"type\": \"number\"\n                  },\n                  \"efficiency\": {\n                    \"type\": \"number\",\n                    \"description\": \"Efficiency in percentage, 0-100\"\n                  },\n                  \"powerFactor\": {\n                    \"type\": \"number\"\n                  },\n                  \"inputVoltage\": {\n                    \"enum\": [\n                      \"100-240v\",\n                      \"100-277v\",\n                      \"220-240v\"\n                    ],\n                    \"type\": \"string\"\n                  }\n                }\n              },\n              \"ratings\": {\n                \"type\": \"object\",\n                \"required\": [\n                  \"protection\"\n                ],\n                \"properties\": {\n                  \"protection\": {\n                    \"enum\": [\n                      \"ip65\",\n                      \"ip66\",\n                      \"ip67\"\n                    ],\n                    \"type\": \"string\"\n                  }\n                }\n              },\n              \"physical\": {\n                \"type\": \"object\",\n                \"required\": [\n                  \"material\",\n                  \"mountingType\"\n                ],\n                \"properties\": {\n                  \"weight\": {\n                    \"type\": \"number\"\n                  },\n                  \"material\": {\n                    \"enum\": [\n                      \"aluminum\",\n                      \"steel\",\n                      \"plastic\"\n                    ],\n                    \"type\": \"string\"\n                  },\n                  \"mountingType\": {\n                    \"enum\": [\n                      \"surface\",\n                      \"recessed\",\n                      \"track\"\n                    ],\n                    \"type\": \"string\"\n                  }\n                }\n              }\n            }\n          }\n        }\n      }\n    },\n    \"header\": {\n      \"type\": \"object\",\n      \"required\": [\n        \"from\",\n        \"to\",\n        \"date\",\n        \"projectName\"\n      ],\n      \"properties\": {\n        \"to\": {\n          \"type\": \"object\",\n          \"required\": [\n            \"companyName\",\n            \"reference\",\n            \"location\",\n            \"attention\",\n            \"tel\",\n            \"email\",\n            \"vatNumber\"\n          ],\n          \"properties\": {\n            \"tel\": {\n              \"type\": \"string\"\n            },\n            \"email\": {\n              \"type\": \"string\"\n            },\n            \"location\": {\n              \"type\": \"string\"\n            },\n            \"attention\": {\n              \"type\": \"string\"\n            },\n            \"reference\": {\n              \"type\": \"string\"\n            },\n            \"vatNumber\": {\n              \"type\": \"string\"\n            },\n            \"companyName\": {\n              \"type\": \"string\"\n            }\n          }\n        },\n        \"date\": {\n          \"type\": \"string\"\n        },\n        \"from\": {\n          \"type\": \"object\",\n          \"required\": [\n            \"companyName\",\n            \"reference\",\n            \"location\",\n            \"tel\",\n            \"email\",\n            \"vatNumber\"\n          ],\n          \"properties\": {\n            \"tel\": {\n              \"type\": \"string\"\n            },\n            \"email\": {\n              \"type\": \"string\"\n            },\n            \"location\": {\n              \"type\": \"string\"\n            },\n            \"reference\": {\n              \"type\": \"string\"\n            },\n            \"vatNumber\": {\n              \"type\": \"string\"\n            },\n            \"companyName\": {\n              \"type\": \"string\"\n            }\n          }\n        },\n        \"projectName\": {\n          \"type\": \"string\"\n        }\n      }\n    }\n  }\n}" verbatim, without appending the response schema. The user message holds only an image_url part. Base64-encode the original local image bytes as a canonical data:image/png, data:image/jpeg or data:image/webp URI (normalize jpg to image/jpeg). Use max_completion_tokens 8192, temperature 0 and response_format {type: "json_schema", json_schema: {name: "fields", strict: true, schema}} with the submitted schema unchanged. Parse choices[0].message.content as JSON and fail loudly when it is missing or invalid.

Deliverables
- A typed client wrapper, an application-level function for this task, error handling for timeouts/empty input, and unit tests with a stubbed client.
- Wire it into my existing stack (ask me which framework if unclear) and add a short usage example.

Run

Output

header.date 2025-02-12

items
reference	brand	quantity	watts	efficiency
ASDZPDI5YL	Lutron	50	103	95
FL5GMWQOUQ	Osram	5	188	93

delivery-note.jpg

Read your documents into your schema

Serverless API

We host the models; you call one API.

No idle GPUs, pay for what you use
Fits your stack: SDK, API, CLI, MCP
Self-host the same open-source engine
SOC 2 Type 2, US or EU data residency
Create an account to try Serverless

Get started

Self-hosted (Kubernetes)

Your data stays in your cloud, backed by our engineers.

Terraform modules for your cloud
Apache-2.0, same engine as Serverless
GPU workers scale to zero between jobs
Separate worker pools for each tenant

Agent prompt Deploy SIE to our AWS account with the superlinked/sie/aws Terraform module. Docs: superlinked.com/docs/deployment.

Deploy SIE to our Google Cloud project with the superlinked/sie/google Terraform module. Docs: superlinked.com/docs/deployment.

Deploy SIE to our Azure subscription with the superlinked/sie/azure Terraform module. Docs: superlinked.com/docs/deployment.

Deploy SIE to our Alibaba Cloud account with the superlinked/sie/alicloud Terraform module. Docs: superlinked.com/docs/deployment.

Talk to us Deploy guide

Run locally

The open-source server on your own machine.

NVIDIA GPU, 80 GB
One container, no cluster
Process your data fully offline
Same SDK and model IDs as Serverless

docker run --gpus all -p 8080:8080 ghcr.io/superlinked/sie-server:latest-cuda13-sglang-cu130

Quickstart

Switch to SIE in 5 minutes

from openai import OpenAI
client = OpenAI()
fields = client.chat.completions.create(
    model="gpt-6.1-sol",
    messages=[
        {"role": "system", "content": instruction},
        {"role": "user", "content": [{"type": "image_url",
            "image_url": {"url": f"data:image/png;base64,{page_b64}"}}]},
    ],
    response_format={"type": "json_schema", "json_schema": {
        "name": "fields", "schema": schema, "strict": True}},
).choices[0].message.content

from anthropic import Anthropic
client = Anthropic()
fields = client.messages.create(
    model="claude-sonnet-5-5",
    system=instruction,
    messages=[{"role": "user", "content": [{"type": "image", "source": {
        "type": "base64", "media_type": "image/png", "data": page_b64}}]}],
    output_config={"format": {"type": "json_schema", "schema": schema}},
    max_tokens=8192,
).content[0].text

SIE

import json
from openai import OpenAI
client = OpenAI(base_url="https://api.superlinked.com/v1", api_key=SIE_API_KEY)
fields = client.chat.completions.create(
    model="google/gemma-4-31B-it:h100-96k-hires-out8k-no-spec",
    messages=[
        {"role": "system", "content": f"{instruction}\n\nJSON schema:\n{json.dumps(schema)}"},
        {"role": "user", "content": [{"type": "image_url",
            "image_url": {"url": f"data:image/png;base64,{page_b64}"}}]},
    ],
    response_format={"type": "json_schema", "json_schema": {
        "name": "fields", "schema": schema, "strict": True}},
).choices[0].message.content

import json
from openai import OpenAI
client = OpenAI(base_url="https://api.superlinked.com/v1", api_key=SIE_API_KEY)
fields = client.chat.completions.create(
    model="google/gemma-4-31B-it:h100-96k-hires-out8k-no-spec",
    messages=[
        {"role": "system", "content": f"{instruction}\n\nJSON schema:\n{json.dumps(schema)}"},
        {"role": "user", "content": [{"type": "image_url",
            "image_url": {"url": f"data:image/png;base64,{page_b64}"}}]},
    ],
    response_format={"type": "json_schema", "json_schema": {
        "name": "fields", "schema": schema, "strict": True}},
).choices[0].message.content