Four months ago, we launched Interfaze. It's a new model architecture that outperforms both closed and open-source models on deterministic tasks like OCR, data extraction, web scraping, and classification. These are tasks where developers expect high accuracy, structured output, and reliable automated backend processing.

We initially tackled this by training SLMs for specialized use cases. This approach gave birth to JigsawStack: one model for per hyper-focused task, bundled with the necessary infrastructure.

JigsawStack Feedback

Speaking with customers, we identified three main challenges with JigsawStack:

• Flexibility: Developers expect fully flexible, unstructured inputs via prompts, alongside structured, developer-defined outputs. Currently, multi-step processes require manual setup. For example, extracting and translating text requires two separate API calls and models.

• Integrability: Because JigsawStack uses custom API schemas, integrating and maintaining it with existing AI SDKs (like LangChain) is difficult.

• Research and Cost: Each model requires its own training pipeline, infrastructure, and research efforts.

We spent over 8 months researching and building Interfaze to solve these exact problems. You can read our paper here: https://arxiv.org/abs/2602.04101

The solution: Interfaze

Interfaze outperforms JigsawStack and generalized LLMs in deterministic developer across all modalities: text, vision, audio, and multilingual tasks. Check out our landing page for benchmarks. Beyond better results, Interfaze offers:

• Flexibility: We support the OpenAI schema out of the box. You can prompt it like any LLM, but you get higher-quality output and additional structured metadata standard LLMs can't provide. Also works with any AI SDK like LangChain.

• Multi-turn calls in one request: You can now extract and translate a document in a single prompt while defining your desired output structure.

• More metadata: Additional metadata produced for tasks like OCR, scraping and more. Check out pre-context to learn more.

• Shared Architecture: Research improvements are shared across tasks. For example, enhancing language support for diarization in STT also improves OCR.

• Backward Compatibility: You can get the exact same outputs as JigsawStack directly within Interfaze using Run task.

JigsawStack & Interfaze Updates

JigsawStack isn't going anywhere. Much of the Interfaze infrastructure is built on top of it. For those needing smaller models running on-premise on a single H100, JigsawStack will continue to power them with constant upgrades and fixes.

However, since most self-serve users are now on Interfaze, we are shifting our branding focus to Interfaze across all socials and emails. Our official company entity will remain JigsawStack.

This transition will happen progressively over the next few days with zero downtime or effects on your systems and APIs.

Upcoming Changes:

• Branding updates from JigsawStack to Interfaze.

• Marketing content will be around Interfaze.

• Interfaze support added to open-source applications like our Translation widget and Google Sheets integration.

Our support team, SOC2 compliance, agreements, and licenses remain unchanged across both Interfaze and JigsawStack.

If you have any questions or want to migrate to Interfaze, Feel free to reach out to https://jigsawstack.com/help :)

👥 Join the JigsawStack Community

Have questions or want to show off what you’ve built? Join the JigsawStack developer community on Discord and X/Twitter. Let’s build something amazing together!