
Institute · Sofia, Bulgaria
INSAIT
Bulgarian AI and computer science institute at Sofia University, in partnership with ETH Zurich and EPFL, behind the BgGPT and MamayLM open language models.
Sofia, Bulgaria
- Founded
- 2022
- Type
- Institute
- Based in
- Sofia, Bulgaria
- Key papers
- 1
- Featured projects
- 8
About
INSAIT, the Institute for Computer Science, Artificial Intelligence and Technology, was created by decree of the Bulgarian Council of Ministers in February 2021 and began operations in April 2022. It is hosted at Sofia University and set up in partnership with ETH Zurich, EPFL and the Government of Bulgaria. Its founder and scientific director is Martin Vechev, a professor at ETH Zurich.
The Bulgarian government committed about 85 million euros over ten years, and donors include Google, AWS, DeepMind, SiteGround and VMware. Besides foundation models, the institute runs research in computer vision, robotics, security and formal methods, and spins out companies such as LogicStar AI.
Its open model line includes BgGPT for Bulgarian and MamayLM for Ukrainian, which was developed with ETH Zurich on Gemma 2 9B. In 2024 it also released COMPL-AI, a framework for evaluating models against the EU AI Act.
Mission and ambitionsINSAIT states its mission as achieving excellence in science, research and education, and it applies this to building language technology for under-served European languages.
Known for
- 01
BgGPT, open Bulgarian language models
- 02
MamayLM for Ukrainian
- 03
COMPL-AI EU AI Act evaluation framework
Milestones
- 2021
Institute created by decree
The Bulgarian Council of Ministers formally created INSAIT with ETH Zurich, EPFL and Sofia University.
- 2022
Operations begin
INSAIT officially began operations in Sofia.
- 2023
AI Alliance founding member
Joined the AI Alliance as a founding member.
- 2024
COMPL-AI
Released an evaluation framework for EU AI Act compliance with ETH Zurich and LatticeFlow AI.
- 2024
BgGPT public release
Published the BgGPT open models for Bulgarian.
- 2025
MamayLM
Released the Ukrainian model MamayLM, co-developed with ETH Zurich.
Open source
A framework for training world models with virtual environments, complete with annotated environment dataset ( (79 stars)
[BMVC 2025] Occam’s LGS: An Efficient Approach for Language Gaussian Splatting (73 stars)
Open-source repository (34 stars)
AR actor for specialist policy training (34 stars)
Model: 2,878 downloads, 8 likes.
Model: 1,895 downloads, 5 likes.
Dataset: 3,556 downloads, 4 likes.
Space: 27 likes on the Hub.
Research papers
Notable releases
Openness
Models and code are public on Hugging Face and GitHub; models adapted from third-party bases carry those licences.
People
- Martin VechevFounder and Scientific Director
- Borislav PetrovExecutive Director
- Virginia Vassilevska WilliamsProfessor listed on INSAIT's site
- Andreas KrauseProfessor listed on INSAIT's site
- Srdjan CapkunProfessor listed on INSAIT's site
Listed from the lab's own pages and public profiles.
Contact and community
Contact through the institute website.
Sources
Compiled from the lab's public pages, Hugging Face and GitHub. Last checked 2026-11-17. Something wrong or out of date? Tell us on Discord.



