ML//model

A model is a concrete implementation of an ML architecture: a trained artifact with specific weights, capabilities and release context. The term covers both proprietary families (GPT-4, Claude, Gemini) and open-weight ones (LLaMA, DeepSeek, Qwen, Mistral).

A model is a concrete implementation of an ML architecture: a trained artifact with specific weights, capabilities and release context. The term covers both proprietary families (GPT-4, Claude, Gemini) and open-weight ones (LLaMA, DeepSeek, Qwen, Mistral).

A model is distinct from its architecture (Transformer, MoE) and from its training techniques (Training): it is the architecture together with the data, the training pipeline and the resulting weights. Since 2024 the competitive landscape has set US frontier labs (OpenAI, Anthropic, Google) against Chinese labs (DeepSeek, Moonshot, Alibaba with Qwen) and European open-weight work (Mistral).