Founded in Bulgaria in 2026.
Synora is an independent AI research and engineering company. It builds its own models, and they are open by default.
Who we are
Synora is founder-funded, with no outside equity. It sets its own research agenda and carries its own risk.
The work is at the level of the models: architectures, weights, training and evaluation. Building on models developed elsewhere would be faster. We decided to build the models.
Our research runs in two lines. Models, inside today's paradigm, specialised for cybersecurity, defence, physical systems and enterprise. And Weightless Intelligence, research into intelligence that does not depend on training at scale.
Synora was founded by Krasimir Petrov, who leads it.
Open by default
Models are released with open weights by default. Each release comes with its evaluation and an account of how it was trained.
Research is published as papers, with methods. Experiments and developments are disclosed as they happen.
The exceptions are stated, and few: work whose release would be unsafe, chiefly in defence and security.
Infrastructure
We train on GPU infrastructure in Stockholm and Frankfurt.
The compute is founder-funded, and supported with cloud credits and GPU capacity by the providers we train with.
We use every serious resource available to us, and scale the infrastructure with the work.
Working together
We work with researchers, engineering teams and organisations whose problems need models built for them, not adapted to them.
Research collaboration, joint development and evaluation partnerships are welcome. Write with a specific problem.
Careers
Synora hires researchers and engineers who work on models directly: pre-training, fine-tuning, evaluation, architectures and simulation.
We look for people who have built something that did not exist before, and can show it. No advertised position is needed to write to us.
Contact
For collaboration, careers, press and everything else, see Contact.