[I.Y.P Consulting] First-Ever Open-Source Release of Proprietary AI Model "SVG-1" Starting Wednesday, April 15, 2026

I.Y.P Consulting, Inc. (Headquarters: Chuo-ku, Tokyo; President & CEO: Jin-Hao Cui; hereinafter IYP) announces that it will begin the open-source release of the beta version of its proprietary AI model, "SVG-1," led by the company's CAIO/CTO and Head of the AI Division, Shohei Osawa, a former Project Associate Professor at the University of Tokyo, starting Wednesday, April 15, 2026. This was announced at "NexTech Week 2026 Spring," held at Tokyo Big Sight from the same day.

SVG-1 (Open-Source Beta Version) Accessible on GitHub https://github.com/iyp-ai/svg

SVG™ is a proprietary technology originating from the research "Support Vector Generation: Kernelizing Large Language Models for Efficient Zero-Shot NLP (SVG)," which was accepted at NeurIPS 2025. IYP had previously announced a poster presentation at NeurIPS 2025 and its policy for an open-source release in 2026. This marks the first step in that plan, starting with the beta release of SVG-1.

SVG-1 is a model designed with an architecture that achieves both high efficiency and interpretability even with limited computational resources, positioning it as a foundational technology for future industrial implementation. IYP envisions its use in applications directly linked to decision-making on the manufacturing floor, such as deployment on in-factory IoT devices, real-time processing, quality inspection, and anomaly detection. In previous releases, IYP has presented quality assessment, anomaly detection, and condition optimization in manufacturing as key use cases, and this release further advances their implementation.

The currently released SVG-1 is a version based on the research accepted at NeurIPS, with its functionality limited to natural language classification and the use of OpenAI Embeddings. Meanwhile, the next-generation version, "SVG-2," which will include generation capabilities, is scheduled for release soon and will feature an embedding function that runs on a CPU.

After joining IYP as Executive Officer and CTO, Shohei Osawa has been leading technology development as the Head of the AI Division, bridging the gap between research and practical field implementation. In releases related to NeurIPS 2025, he stated that the SVG research began with the question of "how to create explainable AI within limited computational resources." This open-sourcing effort is an initiative to connect that philosophy to broader societal implementation.

SVG-1 (Open-Source Beta Version) https://github.com/iyp-ai/svg

Background of the Open-Source Release

In recent years, while the use of generative AI has rapidly expanded, challenges have emerged in manufacturing and on-site operations, such as "PoCs (Proofs of Concept) proceed but don't lead to KPI improvements in factories" and "AI is not continuously used in the field." IYP believes that to move beyond mere demonstrations and link AI adoption to tangible results on-site, a lightweight, easy-to-handle technology foundation with real-time capabilities and implementation suitability is crucial.

SVG is a technology that has been researched and developed based on this awareness of the issues. Previous releases have highlighted SVG's ability to achieve efficiency with a small number of parameters and the international attention it has garnered as a foundational technology research from Japan. This open-source release is an effort to widely open up these achievements to the research community and industry.

Future Outlook

Starting with the release of SVG-1, IYP will further promote on-site implementation, focusing on the manufacturing industry. With an eye on use in in-factory IoT devices and edge environments, the company will advance its deployment for applications like real-time inference, quality inspection, and anomaly detection, while also preparing for the release of the more advanced SVG-2. In addition to generation capabilities, SVG-2 is planned to include the release of an embedding function that runs on a CPU.

Comment from Shohei Osawa, Executive Officer, CAIO/CTO, Head of AI Division

"SVG™ is a technology born from the question of how to realize usable AI in the field, even with limited computational resources. Through this open-sourcing of SVG-1, we aim to connect our research findings to manufacturing sites and industrial applications, rather than letting them remain solely in academic papers. Including the upcoming SVG-2, I hope to cultivate a highly implementable AI foundation together with many more developers, researchers, and practitioners."

History & Related Links

2025.9.17 Announcement of New Structure for Entry into the Generative AI Industry https://prtimes.jp/main/html/rd/p/000000002.000169148.html

2025.9.25 World First! Japanese Company Succeeds in Developing Generative AI (LLM) that Eliminates the Need for GPUs. / Preliminary Announcement at an Event in Tokyo on October 10, 2025 https://prtimes.jp/main/html/rd/p/000000009.000169148.html

2025.9.30 Announcement of Exhibition at NexTech Week 2025 Autumn - AI & Artificial Intelligence EXPO <Pre-registration Link Available> https://prtimes.jp/main/html/rd/p/000000010.000169148.html

2025.12.1 Announcement of New AI Division, Holding Career Consultations for 2026/2027 Graduates in San Diego, USA https://prtimes.jp/main/html/rd/p/000000014.000169148.html

2025.12.23 <Exhibition Report> Announcing "SVG" Research Adopted at NeurIPS 2025 — To Be Open-Sourced in 2026 https://prtimes.jp/main/html/rd/p/000000017.000169148.html

2026.4.8 "We Don't Need AI That Can't Be Used in Factories." Introducing AI Business Reform Consulting That "Sees It Through" to Results on the Manufacturing Floor at <NexTech Week 2026 Spring> from April 15 (Wed) - 17 (Fri) https://prtimes.jp/main/html/rd/p/000000019.000169148.html

SVG-1 (Open-Source Beta Version) https://github.com/iyp-ai/svg