In our search for the technology that best suits our client’s needs, we recently conducted a comprehensive benchmark to evaluate several large language models (LLMs) for Generative AI across different linguistic corpora in Spanish. We decided to focus on models in the 7–12 billion parameter range as they offer an ideal balance between language capabilities and cost-effectiveness. As a consequence, these are some of the most popular models among developers and enterprise customers. In particular, they can run on AWS’s G5 instances (which are cheaper than the high-end instances required for larger models) while still delivering competitive performance. Among these, StabilityAI’s Stable LM 2 model (`stabilityai/stablelm-2–12b-chat`) stands out by consistently delivering superior results over similarly sized models.

WhitePaper «Benchmarking LLMs on Custom Spanish Corpus s»