Extract more exact field values than Claude Sonnet 5.5, 90% cheaper

Turn contracts, claims, invoices and reports into records that fit your JSON Schema. SIE, an open-source inference engine, serves Qwen3.8 27B, an open model held to your schema, so each reply is a typed record ready for your database.

Get started Run in your cloud

SIE Qwen3.8 27B

Fields match the source

NHTSA complaint 11764705

The fire spread extremely rapidly, and the vehicle was substantially engulfed within approximately two minutes and completely destroyed. The sudden loss of propulsion created a crash risk, and the rapid fire placed both occupants at risk of serious injury or death. No physical injuries were reported.

{
  "fire": true,
  "injured_people": 0
}

Claude Sonnet 5.5 bills $7,680 per 1M documents

Claude Sonnet 5.5 (thinking off)

22% of values not exact

Tokens billed: 3,287.42 in and 110.54 out a document
List price: $2.00 in and $10.00 out per 1M tokens
Bill for 1M documents: $7,680

GPT-6.1 Sol (medium reasoning)

26% of values not exact

Tokens billed: 1,947.04 in and 84.07 out a document
List price: $2.00 in and $10.00 out per 1M tokens
Bill for 1M documents: $4,734

arXiv:2604.25359 Study of 21 models: schema compliance is near-perfect, value accuracy is not
The Structured Output Benchmark: A Multi-Source Benchmark for Evaluating Structured Output Quality in Large Language Models

Singh, Khurdula, Khemlani and Agarwal
28 April 2026

Most values exactly right, 90% cheaper than Claude Sonnet 5.5

Run this eval

Field values exactly right

500 documents ↑

SIE Qwen3.8 27B

SIE Qwen3.8 27B 83.5%

Claude Sonnet 5.5

Claude Sonnet 5.5 77.2%

GPT-6 Luna

GPT-6 Luna 76.6%

Claude Haiku 5.5

Claude Haiku 5.5 75.9%

GPT-6.1 Sol

GPT-6.1 Sol 73.7%

Cost per 1M documents

USD ↓

GPT-6 Luna

$227 GPT-6 Luna

Claude Haiku 5.5

$385 Claude Haiku 5.5

SIE Qwen3.8 27B

$728 SIE Qwen3.8 27B

GPT-6.1 Sol

GPT-6.1 Sol $4,734

Claude Sonnet 5.5

Claude Sonnet 5.5 $7,680

Evaluation

Tested: 500 documents from the public Structured Output Benchmark
Scored: Exact match to the answer key, per field
Prices: Each model's own recorded tokens at list prices

Send your schema, get typed field values back

import json
from sie_sdk import SIEClient
client = SIEClient(api_key="API keysk-sie-…")
schema = json.loads("""
schema{
  "type": "object",
  "additionalProperties": false,
  "properties": {
    "model_year": {"type": ["integer", "null"]},
    "make": {"type": ["string", "null"]},
    "model": {"type": ["string", "null"]},
    "speed_mph": {"type": ["integer", "null"]},
    "odometer_miles": {"type": ["integer", "null"]},
    "crash": {"type": "boolean"},
    "fire": {"type": "boolean"},
    "injured_people": {"type": "integer", "minimum": 0},
    "airbags": {"type": "string", "enum": ["deployed", "did not deploy", "unknown", "not mentioned"]}
  },
  "required": ["model_year", "make", "model", "speed_mph", "odometer_miles", "crash", "fire", "injured_people", "airbags"]
}
""")
instruction = "instructionFill the vehicle safety complaint record from this NHTSA complaint narrative. Use null when the narrative does not state a value."
messages = [
    {"role": "system", "content": f"{instruction}\n\nRequired JSON schema:\n{json.dumps(schema, indent=2)}"},
    {"role": "user", "content": "textOn September 5, 2026, at approximately 7:18 p.m. Pacific Time, a 2025 Chevrolet Tahoe High Country equipped with the···please associate this complaint with that investigation and evaluate whether an L87 engine failure initiated the fire."},
]
result = client.chat_completions(
    "modelQwen/Qwen3.8-27B-FP8",
    messages,
    max_completion_tokens=2048,
    extra_body={"temperature":0},
    response_format={
        "type": "json_schema",
        "json_schema": {"name": "answer", "strict": True, "schema": schema},
    },
)
print(json.loads(result["choices"][0]["message"]["content"]))

import { SIEClient } from '@superlinked/sie-sdk';

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

const schema = {
  "type": "object",
  "additionalProperties": false,
  "properties": {
    "model_year": {"type": ["integer", "null"]},
    "make": {"type": ["string", "null"]},
    "model": {"type": ["string", "null"]},
    "speed_mph": {"type": ["integer", "null"]},
    "odometer_miles": {"type": ["integer", "null"]},
    "crash": {"type": "boolean"},
    "fire": {"type": "boolean"},
    "injured_people": {"type": "integer", "minimum": 0},
    "airbags": {"type": "string", "enum": ["deployed", "did not deploy", "unknown", "not mentioned"]}
  },
  "required": ["model_year", "make", "model", "speed_mph", "odometer_miles", "crash", "fire", "injured_people", "airbags"]
};
const instruction = "Fill the vehicle safety complaint record from this NHTSA complaint narrative. Use null when the narrative does not state a value.";
const result = await client.chatCompletions({
  model: 'Qwen/Qwen3.8-27B-FP8',
  messages: [
    {
      role: 'system',
      content: `${instruction}\n\nRequired JSON schema:\n${JSON.stringify(schema, null, 2)}`,
    },
    { role: 'user', content: "On September 5, 2026, at approximately 7:18 p.m. Pacific Time, a 2025 Chevrolet Tahoe High Country equipped with the···please associate this complaint with that investigation and evaluate whether an L87 engine failure initiated the fire." },
  ],
  max_completion_tokens: 2048,
  temperature: 0,
  response_format: {
    type: 'json_schema',
    json_schema: { name: 'answer', strict: true, schema },
  },
});
const content = result.choices[0].message.content;
if (typeof content !== 'string') throw new Error('The model returned no JSON content.');
console.log(JSON.parse(content));

curl https://api.superlinked.com/v1/chat/completions \
  -H "Authorization: Bearer sk-sie-…" \
  -H "Content-Type: application/json" \
  -d "{\"model\":\"Qwen/Qwen3.8-27B-FP8\",\"messages\":[{\"role\":\"system\",\"content\":\"Fill the vehicle safety complaint record from this NHTSA complaint narrative. Use null when the narrative does not state a value.\\n\\nRequired JSON schema:\\n{\\n  \\\"type\\\": \\\"object\\\",\\n  \\\"additionalProperties\\\": false,\\n  \\\"properties\\\": {\\n    \\\"model_year\\\": {\\\"type\\\": [\\\"integer\\\", \\\"null\\\"]},\\n    \\\"make\\\": {\\\"type\\\": [\\\"string\\\", \\\"null\\\"]},\\n    \\\"model\\\": {\\\"type\\\": [\\\"string\\\", \\\"null\\\"]},\\n    \\\"speed_mph\\\": {\\\"type\\\": [\\\"integer\\\", \\\"null\\\"]},\\n    \\\"odometer_miles\\\": {\\\"type\\\": [\\\"integer\\\", \\\"null\\\"]},\\n    \\\"crash\\\": {\\\"type\\\": \\\"boolean\\\"},\\n    \\\"fire\\\": {\\\"type\\\": \\\"boolean\\\"},\\n    \\\"injured_people\\\": {\\\"type\\\": \\\"integer\\\", \\\"minimum\\\": 0},\\n    \\\"airbags\\\": {\\\"type\\\": \\\"string\\\", \\\"enum\\\": [\\\"deployed\\\", \\\"did not deploy\\\", \\\"unknown\\\", \\\"not mentioned\\\"]}\\n  },\\n  \\\"required\\\": [\\\"model_year\\\", \\\"make\\\", \\\"model\\\", \\\"speed_mph\\\", \\\"odometer_miles\\\", \\\"crash\\\", \\\"fire\\\", \\\"injured_people\\\", \\\"airbags\\\"]\\n}\"},{\"role\":\"user\",\"content\":\"On September 5, 2026, at approximately 7:18 p.m. Pacific Time, a 2025 Chevrolet Tahoe High Country equipped with the···please associate this complaint with that investigation and evaluate whether an L87 engine failure initiated the fire.\"}],\"max_completion_tokens\":2048,\"temperature\":0,\"response_format\":{\"type\":\"json_schema\",\"json_schema\":{\"name\":\"answer\",\"strict\":true,\"schema\":{\"type\":\"object\",\"additionalProperties\":false,\"properties\":{\"model_year\":{\"type\":[\"integer\",\"null\"]},\"make\":{\"type\":[\"string\",\"null\"]},\"model\":{\"type\":[\"string\",\"null\"]},\"speed_mph\":{\"type\":[\"integer\",\"null\"]},\"odometer_miles\":{\"type\":[\"integer\",\"null\"]},\"crash\":{\"type\":\"boolean\"},\"fire\":{\"type\":\"boolean\"},\"injured_people\":{\"type\":\"integer\",\"minimum\":0},\"airbags\":{\"type\":\"string\",\"enum\":[\"deployed\",\"did not deploy\",\"unknown\",\"not mentioned\"]}},\"required\":[\"model_year\",\"make\",\"model\",\"speed_mph\",\"odometer_miles\",\"crash\",\"fire\",\"injured_people\",\"airbags\"]}}}}"

Build the "Structured output" 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: Qwen/Qwen3.8-27B-FP8 (OpenAI-compatible endpoint: /v1/chat/completions). Keep the model id configurable.

Task
- Input: a block of text.
- Behaviour: return schema-valid JSON extracted from the text
- Send one POST /v1/chat/completions request (SDK: chat_completions / chatCompletions) with a system message holding the instruction, a blank line, "Required JSON schema:" and the schema as JSON indented by two spaces, and the source text as the user message. Use max_completion_tokens 2048, temperature 0 and response_format {type: "json_schema", json_schema: {name: "answer", strict: true, schema}}. 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

{
  "airbags": "not mentioned",
  "crash": false,
  "fire": true,
  "injured_people": 0,
  "make": "Chevrolet",
  "model": "Tahoe High Country",
  "model_year": 2025,
  "odometer_miles": 18000,
  "speed_mph": null
}

Extract your records 90% cheaper than Claude Sonnet 5.5

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
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

import json
from anthropic import Anthropic

client = Anthropic()
reply = client.messages.create(
    model="claude-sonnet-5-5",
    system=f"{instruction}\n\nRequired JSON schema:\n{json.dumps(schema, indent=2)}",
    messages=[{"role": "user", "content": document}],
    max_tokens=2048,
    output_config={"format": {"type": "json_schema", "schema": schema}},
)
record = json.loads(next(b.text for b in reply.content if b.type == "text"))

import json
from openai import OpenAI

client = OpenAI()
messages = [
    {"role": "system", "content": f"{instruction}\n\nRequired JSON schema:\n{json.dumps(schema, indent=2)}"},
    {"role": "user", "content": document},
]
reply = client.chat.completions.create(
    model="gpt-6.1-sol",
    messages=messages,
    max_completion_tokens=2048,
    response_format={
        "type": "json_schema",
        "json_schema": {"name": "answer", "strict": True, "schema": schema},
    },
)
record = json.loads(reply.choices[0].message.content)

SIE

import json
from sie_sdk import SIEClient

client = SIEClient(api_key="sk-sie-…")
messages = [
    {"role": "system", "content": f"{instruction}\n\nRequired JSON schema:\n{json.dumps(schema, indent=2)}"},
    {"role": "user", "content": document},
]
result = client.chat_completions(
    "Qwen/Qwen3.8-27B-FP8",
    messages,
    max_completion_tokens=2048,
    extra_body={"temperature":0},
    response_format={
        "type": "json_schema",
        "json_schema": {"name": "answer", "strict": True, "schema": schema},
    },
)
record = json.loads(result["choices"][0]["message"]["content"])

import json
from sie_sdk import SIEClient

client = SIEClient(api_key="sk-sie-…")
messages = [
    {"role": "system", "content": f"{instruction}\n\nRequired JSON schema:\n{json.dumps(schema, indent=2)}"},
    {"role": "user", "content": document},
]
result = client.chat_completions(
    "Qwen/Qwen3.8-27B-FP8",
    messages,
    max_completion_tokens=2048,
    extra_body={"temperature":0},
    response_format={
        "type": "json_schema",
        "json_schema": {"name": "answer", "strict": True, "schema": schema},
    },
)
record = json.loads(result["choices"][0]["message"]["content"])