Open Source LLM

An open source LLM is a large language model whose trained weights are published, letting anyone download, run, inspect, and modify it rather than only calling a hosted API. Access to the weights means the model can run on a private server or a laptop, which keeps sensitive data in house, removes per-token pricing, and allows customization through fine-tuning or adapters. Meta's Llama family, Mistral's models, Alibaba's Qwen, Google's Gemma, and DeepSeek's releases anchor the ecosystem, and Hugging Face serves as the main distribution hub. Runtimes such as Ollama, llama.cpp, and vLLM make local and self-hosted serving straightforward, and quantized versions shrink models enough to run on consumer hardware. The label is often imprecise. Most of these releases publish weights but not the training data or full training code, and several carry licenses that restrict commercial use or scale, so open weight is the more accurate term for many of them and the license deserves a careful read before shipping. Self-hosting also shifts work rather than removing it, since capacity planning, GPU cost, safety filtering, and version upgrades become the operator's responsibility instead of the provider's.