Sebastian Raschka has published his curated LLM research paper list covering July through December 2025, organized into nine categories: reasoning models, reinforcement learning methods, inference-time scaling, model releases, architectures, efficient training, diffusion-based language models, multimodal and vision-language models, and pre-training datasets.
This is a reference document, not a survey. Raschka skimmed abstracts and read a small fraction in full, but the value is in the organization: subcategories break reasoning alone into training methods, inference-time strategies, and evaluation, making it a navigable index for anyone working in those areas. It ships alongside his separate annual review, 'State of LLMs 2025: Progress, Problems, and Predictions,' published the same day.
The full list is paywalled for paid subscribers of Ahead of AI. What makes it worth reading is not any single paper but the taxonomy itself, which reflects where active research actually concentrated in the second half of 2025. If you want to know what problems the field was attacking, the category structure tells you before you open a single PDF.
[READ ORIGINAL →]