8 Types of LLMs Used in AI Agents Explained for Engineers

ayushthakur0 · x · 2026-08-29

A breakdown of 8 essential LLM architectures for Gen AI Data Scientists and AI Engineers:

GPT (Generative Pretrained Transformer): General-purpose text understanding and generation backbone.
MoE (Mixture of Experts): Routes tokens to specialized “experts” to scale capacity efficiently.
LRM (Large Reasoning Model): Tuned for multi-step reasoning and RAG/tool use.
VLM (Vision-Language Model): Processes images + text for multimodal perception and grounding.
SLM (Small Language Model): Compact, fast models for edge/on-device or low-latency tasks.
LAM (Large Action Model): Plans and executes actions via tools/APIs/robots (agent actuation).
HLM (Hierarchical Language Model): Layered coordination (e.g., user/item/task submodels) for complex workflows.
LCM (Large Concept Model): Maps words to higher-level concepts for abstraction and semantic understanding.

More from Models