AMD Silo AI has released two new open-source models in its Poro 2 large language model lineup: Poro 2 8B Long and Poro 2 8B Math Reasoning. Poro 2 8B Long features a 16x expansion of its context window to 131,072 tokens, a 16-fold increase over the original 8,192-token limit, enhancing utility in Retrieval Augmented Generation workflows for industries like legal, academic, or enterprise search. Poro 2 8B Math Reasoning is a math-focused variant designed for reasoning in Finnish and English, addressing a gap in multilingual reasoning models, though it remains in preview and not production-ready. Both models and associated training resources are available on Hugging Face. The Poro family, developed with TurkuNLP and trained on the LUMI supercomputer, builds on the methodology introduced with Poro 34B in 2024. The original Poro 2 models are based on Meta’s Llama 3.1 architecture. Both new models are released under open-source licenses, along with playbooks and training datasets detailing their development on AMD Instinct GPUs, aligning with the goals of the OpenEuroLLM consortium to strengthen digital sovereignty.