DFM Mimir v1: An Open HRM Delivering Frontier Performance at 1B Parameters Using Only Permissible Post-Training Data
A 1B-parameter HRM model is being presented as competitive with larger frontier models while using only permissible post-training data.
The paper says Mimir v1 was trained from scratch on a mixture of 161 datasets. Its authors report gains over the original HRM-Text 1B and comparisons with Qwen 3.5 4B and Gemma 4 E2B across 20 benchmarks. They claim especially strong Danish results, including a new state of the art. The model is available on Hugging Face. Source: ArXiv · AI/CL/LG's note.
The paper says Mimir v1 was trained from scratch on a mixture of 161 datasets. Its authors report gains over the original HRM-Text 1B and comparisons with Qwen 3.5 4B and Gemma 4 E2B across 20 benchmarks. They claim especially strong Danish results, including a new state of the art. The model is available on Hugging Face. Source: ArXiv · AI/CL/LG's note.
score 6