Glossary Model Categories
Model Categories
Also known as:
llm categories, ai model categories, model strengthsAIAgentic
Definition
The usual ways AI models are grouped and compared: the job a vendor positions a model for, and the axes benchmark leaderboards rank them on
A model usually sits in several at once, and vendors lead with one. The categories describe positioning, not a guarantee: the useful test is still the task in front of you.
- General-purpose (
general): Balanced performance across tasks: the default choice when nothing more specific applies. - Coding (
coding): Code generation, debugging and software engineering, including working in a repository as an agent. - Reasoning (
reasoning): Logical and mathematical problem solving and exam-style reasoning, often with extended thinking before answering. - Multilingual (
multilingual): Question answering and translation across languages. - Visual reasoning (
visual-reasoning): Interpreting and reasoning over images: charts, screenshots, diagrams, photos. - Agentic (
agentic): Multi-step, autonomous task planning and execution with tools: the loop a harness runs. - Specialized (
specialized): A narrower job, such as writing, research, long context, speech synthesis or transcription.
The tech registry uses the same keys, as each model's modelCategories. See the
AI models and model.