DFM Mimir v1: 1B parameter model trained on ethical data achieves Danish SOTA

A new 1-billion-parameter language model named Mimir v1 has been developed, utilizing a Hierarchical Reasoning Model (HRM) architecture. This model is notable for being trained exclusively on permissible data, setting a new state-of-the-art for Danish language performance while remaining competitive in English. Mimir v1 was trained on a mix of 161 datasets and demonstrates performance comparable to larger models like Qwen 3.5-4B and Gemma 4-E2B. AI

IMPACT Demonstrates that ethical data sourcing can yield competitive LLM performance, potentially lowering barriers for open-source development.

RANK_REASON The cluster describes a new research paper detailing a novel language model architecture and its performance. [lever_c_demoted from research: ic=1 ai=1.0]

Read on Hugging Face Daily Papers →

• DFM Mimir v1
• Gemma 4-E2B
• Hierarchical Reasoning Model
• Hugging Face Hub
• Qwen 3.5-4B

• model release
• paper

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →