SteuerLLM: An Open-Source AI Model for German Tax Law

An international team of researchers has released SteuerLLM, am open-source, 28-billion-parameter AI model trained on German tax law materials. In a companion article, the team reports that SteuerLLM outperforms general-purpose instruction-tuned models of comparable size and, in some tests, substantially larger systems. These results suggest that domain-specific design may matter more than parameter count for specialized legal reasoning.

At the very least, these findings suggest a central role for domain-specific AI models in legal practice, with bigger questions about whether the optimal domain is “law” or “subfields within the field of law.” Or both. Or, with generalist AI models, all three. Links to the web demo, the paper, and more, below the fold.

SteuerLLM is notable because its model is unusually open, with details released about weighting, benchmarks, and construction. Moreover, SteuerLLM runs against commercial trends towards general-purpose legal AI platforms such as Harvey and Legora. SteuerLLM is smaller than mainstream models—and likely more resource-efficient—without sacrificing performance on evaluation tasks.

SteuerLLM’s web demo is available at https://steuerllm.i5.ai.fau.de/. The demo accepts input in English; output appears in German. Users may, of course, use other AI tools to translate this output. After some light testing, I’m impressed by SteuerLLM’s depth and accessibility. Although the structure of tax law may lend itself to SteuerLLM’s tightly domain-specific approach, the model’s performance suggests that the legal profession should continue testing (and probably deploying) AI technology across multiple registers.

The bigger question is not just whether lawyers should use specialized AI models. It’s also about the level at which specialization should occur. The full article on SteuerLLM:

Sebastian Wind, Jeta Sopa, Laurin Schmid, Quirin Jackl, Sebastian Kiefer, Fei Wu, Martin Mayr, Harald Köstler, Gerhard Wellein, Andreas Maier & Soroosh Tayebi Arasteh, ⁠SteuerLLM: Local Specialized Large Language Model for German Tax Law Analysis, 16 Scientific Reports (2026):

Large language models (LLMs) demonstrate strong general reasoning and language understanding, yet their performance degrades in domains governed by strict formal rules, precise terminology, and legally binding structure. Tax law exemplifies these challenges, as correct answers require exact statutory citation, structured legal argumentation, and numerical accuracy under rigid grading schemes. We algorithmically generate SteuerEx, the first open benchmark derived from authentic German university tax law examinations. SteuerEx comprises 115 expert-validated examination questions spanning six core tax law domains and multiple academic levels, and employs a statement-level, partial-credit evaluation framework that closely mirrors real examination practice. We further present SteuerLLM, a domain-adapted LLM for German tax law trained on a large-scale synthetic dataset generated from authentic examination material using a controlled retrieval-augmented pipeline. SteuerLLM (28B parameters) consistently outperforms general-purpose instruction-tuned models of comparable size and, in several cases, substantially larger systems, demonstrating that domain-specific data and architectural adaptation are more decisive than parameter scale for performance on realistic legal reasoning tasks. All benchmark data, training datasets, model weights, and evaluation code are released openly to support reproducible research in domain-specific legal artificial intelligence.

TaxProf Blog

About the Author

Stanford Launches a “Legal Engineering Academy”

A Beckerian Theory of Taxation

Will JD Hiring Plunge as Enrollment Booms?