Llama

Placed against the broader landscape of large language models, Meta's family became the de-facto foundation of the open-source AI ecosystem, the base that much of the field builds on. Available from a few billion to hundreds of billions of parameters under a permissive commercial licence, these models can be fine-tuned and self-hosted, and they power countless products through inference providers like Groq, Together and Fireworks. A startup that wants to fine-tune a model on its own data and serve it cheaply at high volume fits the pattern: pick a size, tune it, and run it via a low-cost provider or its own hardware for coding, research, data-analysis and content-creation. Multimodal variants extend the range beyond text. Positioning is a matter of ecosystem scale. Unlike Gemini, a closed commercial assistant, this ships open weights you can host and customise. Against fellow open contenders Mistral, Qwen and DeepSeek, its standout is the massive community and tooling built around it, plus very low inference cost and availability across many providers. The costs are practical: self-hosting requires infrastructure, and raw models need tuning for production rather than working polished out of the box. It suits teams that want a customisable, self-hostable base with broad provider support. Choose it for ecosystem breadth, low inference cost and fine-tuning freedom; look elsewhere if you lack the infrastructure to host or want a ready-to-use hosted assistant.