OLMo 2 13B Instruct

OLMo 2 13B Instruct is Allen AI's fully-open 13B parameter language model, released November 2024 with training data, weights, and training code all published under Apache 2.0. That full-stack openness — Dolma corpus, OLMo-mix training framework, public eval results — makes it the most reproducible model in the sub-20B class, which is why academic teams and ML researchers consistently reach for it over alternatives. The hard constraint for production work is context: 4K tokens rules out most RAG architectures without chunking. Hosted coverage is thin; you're more likely running this on your own GPU cluster than pulling it from a managed endpoint. Worth tracking as a fine-tuning base if reproducibility or auditability requirements are non-negotiable.