Backed by
Combinator
A model built for deterministic tasks
Accurate, verifiable output in one call.
A new model architecture for OCR, audio understanding, structured data extraction and more, built for the highest accuracy, precision and consistency.
Drop a PDF or image, or click to browse
One file, up to 20MB
Try one:
Verifying your browser before you can run the demo...
Trusted by builders at
- Deterministic accuracy
- Consistent, high-precision results for repeatable production tasks.
- Verifiable outputs
- Confidence scores and bounding boxes you can build rules on.
- Multimodal input
- Text, images, audio, files and video in a single model.
- Multilingual by default
- Work across languages with built-in understanding and translation.
- Controllable
- Configure parameters and guardrails to fit your workflow.
- Intelligence
- Reasoning for complex instructions and understanding tasks.
Works with the SDK you already use
Official TypeScript and Python SDKs, and OpenAI compatible, so the OpenAI SDK, Vercel AI SDK and LangChain work unchanged.
Interfaze SDK
Vercel AI SDK
LangChain SDK
import { Interfaze, responseFormat } from "interfaze";
import { z } from "zod";
const interfaze = new Interfaze({
apiKey: "<your-api-key>"
});
const IDSchema = z.object({
first_name: z.string().describe("First name on the ID"),
last_name: z.string().describe("Last name on the ID"),
dob: z.string().describe("Date of birth on the ID"),
driver_licence_number: z.string().describe("Driver licence number on the ID"),
});
const response = await interfaze.chat.completions.create({
messages: [
{
role: "user",
content: [
{ type: "text", text: "Extract the details from this ID" },
{
type: "image_url",
image_url: {
url: "https://r2public.jigsawstack.com/interfaze/examples/id.jpg",
},
},
],
},
],
response_format: responseFormat(z.toJSONSchema(IDSchema), "id_schema"),
});
console.log(JSON.parse(response.choices[0]?.message.content ?? "{}"));Two models, one API
Interfaze 1 for the highest accuracy, Interfaze 1.1 Lite with open weights you can self-host.
Our flagship model for the highest accuracy on deterministic tasks.
Input
/ MTok
Output
/ MTok
Context window
1M tokens
Max output
32k tokens
Reasoning
Available
Input
Text · Images · Audio · File · Video
Smaller open weight model. Available for self-hosting.
Input
/ MTok
Output
/ MTok
Context window
128k tokens
Max output
32k tokens
Reasoning
Available
Input
Text · Images · Audio · File
How it compares
Public benchmarks against models in the same price tier.
Overview
OCRBench V2
olmOCR
RefCOCO
VoxPopuli-Cleaned-AA
SOB Value Acc
Spider-2.0-Lite
GPQA Diamond
MMMLU
MMMU-Pro
Radar chart
| Benchmark | Interfaze-1 | Interfaze-1.1-Lite | Gemini-3.8-Flash | Claude-Sonnet-5 | GPT-5.4-Mini | Grok-4.3 |
|---|---|---|---|---|---|---|
OCRBench V2 Native OCR | 70.7% | 60.9% | 63.9% | 59.2% | 52.7% | 54.7% |
olmOCR Complex document processing | 85.7% | 83.8% | 82.3% | 83.5% | 80.1% | 81.9% |
RefCOCO Object detection (NL prompts) | 82.1% | 83.8% | 80.9% | 69.2% | 67.0% | 25.0% |
VoxPopuli-Cleaned-AA ASR (speech recognition) | 2.4% | 3.0% | 4.0% | — | — | — |
SOB Value Acc Structured output | 80.5% | 81.5% | 80.2% | 79.3% | 75.1% | 78.4% |
Spider-2.0-Lite Text-to-SQL | 52.9% | 48.9% | 46.7% | 50.6% | 26.7% | 45.9% |
GPQA Diamond PhD-level problem solving | 92.4% | 85.9% | 91.4% | 78.3% | 82.8% | 73.6% |
MMMLU Multilingual Q&A | 90.9% | 87.8% | 88.1% | 87.3% | 75.3% | 89.7% |
MMMU-Pro Multimodal understanding | 71.1% | 73.2% | 64.2% | 53.3% | 40.4% | 68.7% |
*Bold cells mark the leader on each benchmark. '—' means the model doesn't support that modality.
Every value comes back with its box and confidence
Check any field against the page, and send low-confidence values to review instead of trusting them blindly.
100+ languages, no translation step
Text, audio and images in over 100 languages, read and translated in the same call.
One call instead of a pipeline
Documents and audio in, structured output back. No OCR tool, speech API, classifier, LLMs and glue code to wire together.
Browsing and code execution, built in
Headless browsers and sandboxes run inside the call, so there is nothing extra to host or maintain.
Guardrails you set per request
Turn safety categories on or off for text and images on each request. A blocked request comes back flagged unsafe.
Specialised models, one transformer
A Mixture-of-Architecture model: task-specific DNNs and CNNs for OCR, detection, speech and more, joined by a transformer layer. The precision of specialised models with the flexibility of an LLM.
Latest blog posts
FAQs
Start building with Interfaze.
Deterministic accuracy for your workflows.
Start for free. No credit card required.