German Sovereign Model Just Tweaks Data Mix
eliebakouch · x · 2026-07-12
This comment reviews a "sovereign" foundation model released by a German consortium:
- Its architecture and most hyperparameters are largely borrowed from Nemotron 3 Nano.
- The primary change is a roughly 20% adjustment in the data mixture.
- Training scale reached approximately 26T tokens and was completed on domestic clusters.
Thus, its "sovereignty" is rooted in training infrastructure and data/deployment control rather than architectural innovation. The author also questions the computational efficiency of this approach, suggesting that continued pre-training might have yielded comparable or even superior results.
Related event: German Alliance Releases Open-Source Sovereign Model Soofi S 30B-A3B(5 posts)→
More from Models
- GPT-6.1 Terra lands: early users praise speed, value and subagent performance — BenBajarin · 2026-10-27
- Leaked GPT-6.1 Terra benchmarks: cheap but 'boring', still far behind Astra and Claude — Angaisb_ · 2026-10-27
- Artificial Analysis breaks down all sub-evaluations in Intelligence Index v4.3.2 — ArtificialAnlys · 2026-10-27
- GPT-6.1 Terra gains ~20 points on Terminal-Bench 4.0, rivalling GPT-6.1 Sol — ArtificialAnlys · 2026-10-27
- GPT-6.1 Terra cuts hallucination rate by 50 points while lifting accuracy 2 points — ArtificialAnlys · 2026-10-27
- GPT-6.1 Terra scores 48 on AA Coding Agent Index at one-third the cost of GPT-5.6 Terra — ArtificialAnlys · 2026-10-27