---
title: "DFM Mimir v1: An Open HRM Delivering Frontier Performance at 1B Parameters Using Only Permissible Post-Training Data"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2608.13517v1"
permalink: "https://feed.astralor.com/items/itm_92581d00b3c6996b2f7c16c3"
published: "2026-08-13T17:37:53.000Z"
score: 75
channel: "ai"
tags: ["Danish Foundation Models","DFM Mimir v1","HRM","Permissible Data","Open-source Model"]
---

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

## AI commentary

> This paper is worth watching because it proves that small, clean-data models can rival models several times their size, especially for teams with limited compute or strict data-licensing constraints. If you're exploring low-cost multilingual or Danish scenarios, Mimir v1 may be more practical than much larger counterparts.

## At a glance

DFM Mimir v1 is a 1-billion-parameter language model based on the Hierarchical Reasoning Model (HRM) architecture, trained from scratch using only permissible post-training data. It delivers highly competitive performance in English and sets a new state of the art for Danish, outperforming HRM-Text 1B and competing with larger models such as Qwen 3.5 4B and Gemma 4 E2B across 20 benchmarks covering English, Math & Code, and Danish. The model is publicly available on the Hugging Face Hub, trained on a mixture of 161 datasets. This work offers a high-quality open-source alternative for researchers constrained by data-permissibility requirements.

## Key points

- DFM Mimir v1 has only 1B parameters, is based on the HRM architecture, trained from scratch with only permissible post-training data.
- Across 20 benchmarks for English, Math & Code, and Danish, it outperforms HRM-Text 1B and competes with larger models.
- Trained on 161 datasets, the model is open-sourced on the Hugging Face Hub.

## Full text

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. We introduce Mimir v1, a 1-billion-parameter language model based on the Hierarchical Reasoning Model (HRM) architecture, that is trained from scratch and delivers highly competitive performance for English and sets a new state of the art for Danish using only permissible post-training data. Trained on a mixture of 161 datasets, Mimir v1 outperforms the original HRM-Text 1B and competes with larger frontier models like Qwen 3.5 4B and Gemma 4 E2B, tested across 20 benchmarks for English, Math & Code and Danish. The model is available on the Hugging Face Hub: https://huggingface.co/danish-foundation-models/DFM-Mimir

---

Verify against the original before citing; curated by Astral Feed: https://feed.astralor.com/items/itm_92581d00b3c6996b2f7c16c3