Glossary
AI subscription terms, explained
The jargon behind the scores, in plain English.
| Context window | How much text (in tokens) a model can consider at once. Bigger windows handle longer documents and chats. ~750 words ≈ 1,000 tokens. |
|---|---|
| Token | The unit models read and write – roughly a short word or word-piece. Limits and context are measured in tokens. |
| Frontier model | The newest, most capable model a lab offers. Paid plans usually unlock it; free tiers may not. |
| Reasoning model | A model tuned to think step-by-step for harder maths, logic and coding tasks. |
| Agentic / tool use | When the AI can take multi-step actions and call tools (search, code, browse) to complete a task, not just answer. |
| Multimodal | Understands more than text – images, audio, sometimes video. |
| RAG | Retrieval-augmented generation: the AI looks up sources first, then answers – the basis of cited, research-style answers. |
| Hallucination | When a model states something false confidently. Always verify important facts. |
| Aggregator / all-in-one | A subscription that bundles several models behind one interface so you can compare or switch. |
| Token limit / rate limit | Caps on how much you can use per period on a given plan. |
| SSO / SCIM | Single sign-on and user provisioning – important for teams and enterprises. |
| Zero-data-retention | An option where prompts aren't stored after processing – used in sensitive/enterprise settings. |