IAAIMimir
Mimir: A Sub-2B Hierarchical Reasoning Model Trained on Permissible Data Punches Above Its Weight Class
A 1.78B language model built on the HRM-Text architecture, whose layers form a low-level and a high-level stack applied repeatedly in an interleaved schedule, trained from scratch only on permissibly licensed data.
Inside the method
[ L-stack × n ↔ H-stack ] × m (interleaved)
Simplified conceptual schematic. Consult the paper for the complete architecture.
- Recurrence family
- Full-stack recurrence
- Depth control
- Fixed schedule; L-stack applications adjustable at inference
- KV / state strategy
- See paper
Reading note
Mimir v1.5 reports the best sub-2B average across 20 English, Danish and math/code benchmarks, ahead of HRM-Text 1B and Ouro 1.4B/2.6B. Changing L-stack applications degrades gracefully while changing H-stack applications breaks the model.
Sources checked 2026-10-28. This catalog does not imply independent reproduction.
Related works
Cite this work
@misc{mimir2026,
title = {Mimir: A Sub-2B Hierarchical Reasoning Model Trained on Permissible Data Punches Above Its Weight Class},
author = {Peter Schneider-Kamp and Jacob Nielsen and Gianluca Barmina and Danial Namazifard and Annemette Brok Pirchert and Kenneth Enevoldsen and Lukas Galke Poech},
year = {2026},
eprint = {2608.13517},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2608.13517}
}