6. Language Models

Foundation models specialized to text/tokens. Two child axes: lineage (statistical → neural → SLM/LLM — a capability/scale progression) and post-training role (base → instruct → chat → code → reasoning). Base models predict tokens; the others are the same engine shaped by training toward a use. This is the cultural center of gravity of the field.

Children

• statistical language models
• neural language models
• small language models / SLM
• large language models / LLM
• base models
• instruct models
• chat models
• code models
• reasoning models
• multimodal LLMs

Related

• Foundation Models — the broader class
• Transformer Architecture — what LLMs are built from
• Training & Post-Training — what turns a base model into instruct/chat
• Reasoning & Test-Time Compute — reasoning models