AI research & engineering lab in Austria building the next generation of intelligent systems.
We have proven our efficiency and quality techniques by matching major labs’ nano-class models with noeum-1-nano —a model trained entirely from scratch. No pretrained weights, no inherited shortcuts. Self-funded across multiple pretraining and post-training stages. Now, we’re ready to multiply that output. With the right compute infrastructure and backing, we can scale up and compete directly with the largest AI labs. Our name reflects our mission: Noeum derives from “mind,” “meaning,” and “thought.” We’re focused on building genuine reasoning systems.
noeum-1-nano
We recently completed noeum-1-nano — a 0.6B parameter MoE model (0.2B active) that performs competitively with models from major labs at similar scale and training compute. Built entirely from scratch: custom tokenizer, pre-training pipeline, and full post-training stack including SFT, GRPO, and DPO.
LLMification
Transforming human knowledge from readable to learnable. Books worked problems become SFT examples, practice problems become RL environments with verifiable answers, and synthetic generators create infinite variations from single templates.