Models

Interfaze comes in two models.

• interfaze-1 is the flagship. It has the highest accuracy and a 1M-token context window.
• interfaze-1-lite is a smaller any-to-text model with open weights under the Apache 2.0 license. It has a 128k-token context window, takes text, images, audio and files, and runs on a single 80 GB GPU, so you can self-host it.

Which model should I use?

Pick interfaze-1 when:

• accuracy matters most, for example OCR on dense or degraded documents
• your input is long: big PDFs, long transcripts or many files in one request

Pick interfaze-1-lite when:

• you want to self-host the model, or run it offline
• you want a smaller model for OCR, transcription, extraction or classification
• your input fits in 128k tokens

Lite matches or beats interfaze-1 on multimodal reasoning (MMMU-Pro), visual grounding (RefCOCO) and structured output (SOB). interfaze-1 leads on text in images (OCRBench V2), science (GPQA Diamond), multilingual knowledge (MMMLU) and text-to-SQL.

Compare specs

View pricing | View limits | View leaderboard

Switch models

Set the model param on any request. Everything else stays the same.

import { Interfaze } from "interfaze";

const interfaze = new Interfaze();

const response = await interfaze.chat.completions.create({
	model: "interfaze-1-lite",
	messages: [{ role: "user", content: "Summarise the attached contract in three bullets." }],
});

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

When you don't set model, the Interfaze SDK and LangChain integration run on interfaze-1.

Model cards

• interfaze-1: specs, benchmarks, capabilities and architecture.
• interfaze-1-lite: specs, benchmarks, capabilities, architecture and how to self-host.