DFM Mimir v1: An Open HRM Delivering Frontier Performance at 1B Parameters Using Only Permissible Post-Training Data

Current large language model development relies on massive, often non-permissible datasets, creating a high barrier for researchers committed to open-source and ethically sourced data. Mimir v1 addresses this by delivering competitive performance using only permissible post-training data.

Mimir v1 is a 1-billion-parameter model based on the Hierarchical Reasoning Model (HRM) architecture, trained from scratch on a mixture of 161 datasets. It focuses on English and Danish, aiming to set a new state of the art for Danish while maintaining strong English performance.

Across 20 benchmarks spanning English, Math & Code, and Danish, Mimir v1 outperforms the original HRM-Text 1B and competes with larger frontier models like Qwen 3.5 4B and Gemma 4 E2B. It achieves a new state of the art for Danish, demonstrating that permissible data can yield frontier-level results at a smaller scale.

The model is openly available on the Hugging Face Hub, enabling researchers to build on ethically sourced, open models. This work lowers the barrier for open-source and ethically conscious AI development.