german

The paper introduces Alice, a large-scale German rubric-based benchmark for automatic short answer scoring, covering three subtasks: learning performance (Alice-LP), knowledge elements (Alice-KE), and skills (Alice-SK). It compares sequence classification, zero-shot LLM prompting, and rubric-retrieval, finding the latter most consistent on the novel fine-grained subtasks while zero-shot LLMs lag behind supervised methods.

This paper investigates whether forcing LLMs to reason in the same language as the query and retrieved documents improves monolingual non-English RAG performance. Using a German RAG testbed built over the fictional world of The Dark Eye, the authors find that aligning the reasoning language with the evidence helps (German reasoning beats French despite the model's stronger French proficiency), but native unconstrained English reasoning still matches the aligned setting, highlighting the need for truly multilingual reasoning models.

Aleph Alpha released Kolibri, a sovereign open-weight English-German Mixture-of-Experts model with 78B total and 3B active parameters, 1M-token context, and Apache 2.0 licensing, aimed at regulated mission-critical deployments in public administration, industry, and aerospace.

The author is highly impressed with Qwen 3.8 27B's German translation capabilities, noting its superior vocabulary, sentence structure, and punctuation compared to other frontier models like GPT-5.6 and Fable 5.

This paper introduces AVShift, the first German benchmark for authorship verification under distribution shifts in genre, time, and AI-era. It evaluates feature-based, embedding-based, and LLM-based approaches, finding that temporal drift significantly impacts performance while no measurable AI-era shift is detected.

A user shares their positive experience with Gemma 4 26b A4b, praising its speed, multimodal ability, and strong language capabilities, especially in German, while noting it handles various tasks well despite not matching Qwen's agentic and coding performance.

This paper introduces GEMCo, the first publicly available German corpus of multi-turn e-mail counselling conversations, validated against real counselling data to serve as an ethically releasable proxy for inaccessible sensitive data.

German AI consortium releases Soofi S, an open 30B mixture-of-experts model that achieves top scores on English and German benchmarks, using a hybrid Mamba-Transformer architecture for efficient long-context inference.

The Soofi S 30B-A3B is a sovereign, open-source Mixture-of-Experts foundation model for German and English, pretrained on 27 trillion tokens. It matches dense 14-27B models on benchmarks while offering superior long-context throughput and outperforms all European open baselines.

SoofiS30B-A3B is a sovereign, open-source Mixture-of-Experts hybrid Mamba Transformer foundation model for German and English, activating only 3B of 30B parameters per token and achieving strong benchmark performance.

Introduces Schützen, a safety dataset for evaluating LLMs in Bulgarian and German, revealing cross-language differences in safety behavior and advocating for region-specific evaluation resources.

This paper presents a transformer-based model that projects political orientation of German texts onto a continuous left-to-right spectrum, achieving high accuracy across multiple corpora including Bundestag plenary notes, Wahl-O-Mat, newspapers, and tweets.