Large language models — the full vocabulary
Every term you'll encounter using an LLM API, evaluating a chat model, or explaining how ChatGPT works. From tokens to context windows to streaming — the working vocabulary of the LLM era.
Large language models are the dominant AI paradigm of the current era. Every conversation with ChatGPT, Claude, or Perplexity, every API call to GPT-5 or Claude Opus 4.8, every fine-tuned specialist model — they all share the vocabulary in this category. Terms here are the day-to-day language of anyone building with LLMs.
The 100 terms are grouped into five sub-topics: how LLMs work at a high level, how you prompt them, how sampling produces the output you see, how context management works, and the operational vocabulary you need in production. Some terms overlap with deep learning architectures (where transformer lives) and prompt engineering (deeper prompting techniques).
The most important terms in LLMs & Language Models
Start here if you're new. These entries explain the foundational vocabulary in depth.
Large Language Model (LLM)
A transformer-based model trained on massive text corpora to predict the next token. Powers ChatGPT, Claude, Gemini, and essentially every conversational AI product.
Read the full entry →ContextContext Window
The maximum number of tokens (input plus output) a model can process in a single conversation. Modern LLMs range from 8K to 2M+ tokens.
Read the full entry →FoundationsToken
The unit of text a language model actually processes — roughly 3-4 characters or 0.75 words in English. Tokens determine cost, context math, and edge cases.
Read the full entry →SamplingTemperature
A number (usually 0-2) that controls output randomness. 0 for deterministic, 0.7-1.0 for creative tasks. Different providers implement it slightly differently.
Read the full entry →PromptingSystem Prompt
The instructions that shape the model's behavior before user messages arrive. Sets persona, constraints, and defaults.
Read the full entry →OperationsStreaming
Delivering the response token-by-token as the model generates rather than waiting for completion. Standard for chat UIs.
Read the full entry →All LLMs & Language Models terms, organized
Grouped into sub-topics so you can find neighbors and prerequisites, not just alphabetical entries.
LLM foundations
Prompting & messages
Sampling & generation
Context management
Operations & production
Related term categories
These categories connect naturally to LLMs & Language Models — many terms cross-reference between them.
Go deeper on LLMs & Language Models
📖 Related concept tutorials
Long-form guides that walk through how these concepts actually work.
- How Transformers Work
The architecture behind every LLM.
- Understanding Context Windows
Why longer isn't always better.
- What is a Token
Tokenization explained with real examples.
🎯 Related models & tools
Real products and models where you'll encounter these terms.
- Claude Opus 4.8
Anthropic's flagship LLM.
- GPT-5
OpenAI's flagship LLM.
- Gemini 3
Google's flagship LLM.
Questions about LLMs & Language Models
No, but they dominate. Smaller specialized language models exist for classification, translation, and domain-specific tasks. But when people say 'language model' in 2026, they almost always mean an LLM.
Because tokens determine cost (per-million pricing), context math (how much fits in the window), latency (longer input = slower first token), and rate limits. Nearly every practical LLM decision comes back to tokens.
They're often used interchangeably. Strictly, context window is the maximum the model supports; context length is how much you're actually using. In practice, both refer to the token capacity you're working within.
Not exactly — it means more random. High temperature can produce creative results, but also incoherence and hallucinations. For factual tasks, low temperature (0-0.3) is usually right.
A base model just predicts the next token given text. A chat model has been fine-tuned to follow instructions and respond in a helpful, harmless conversational format. Almost all production APIs expose chat models.