Language models
We train and evaluate our own language models, including conventional and experimental architectures.
MatrixLM · Brussels, Belgium
AI models,
built in Belgium.
MatrixLM is an independent AI lab in Brussels developing and training language models and the infrastructure required to make them useful.
We build our own models, experiment with new model architectures, and develop systems for conversational products and developer access.
Developer infrastructure
We are building the inference and API infrastructure required to make our models accessible to developers and applications.
Architecture R&D
We experiment with new approaches to language-model architecture, training, retrieval, and inference.
The lab
Building models from the ground up.
MatrixLM started as an independent AI project in Brussels, built around the idea that useful AI models can be developed locally rather than relying exclusively on models and infrastructure built elsewhere.
Our work spans model training, evaluation, retrieval, inference, developer infrastructure, and experimental neural architectures.
The goal is to turn that work into AI products and model infrastructure that people and developers can use directly.
Research & development
Training today.
Experimenting beyond it.
We train conventional language models while also investigating new approaches to model architecture and computation.
Experimental systems are evaluated through controlled comparisons covering training behavior, retrieval, generalization, reasoning, and generation.
Some of this work involves experimental architectures that are not yet publicly documented.
~10B
Training tokens
completed in a current model run
Multiple
Model approaches
under active development
Research and development is ongoing. Public figures describe internal development work and are not third-party benchmarks.
Belgian AI with European ambition.
MatrixLM is being developed in Brussels, Belgium. Our aim is to make capable AI developed here available to people, developers, and businesses.
Over time, we want to contribute to a stronger European AI ecosystem through locally developed models, research, and infrastructure.