← Back to Home

Glossary

Every technical term on Antbase, explained in plain language. If something on the site confuses you, you'll find the answer here.

ABCEFILMNOPQRSTVW
AI BasicsContext & LimitsModel CapabilitiesAntbase & ANTPricing & BillingAPI & IntegrationAdvanced Concepts

AI Basics

AI Model
A program that has been trained on huge amounts of text so it can read your messages and write helpful replies. Think of it like an extremely well-read assistant.
Browse 16,000+ models
LLM (Large Language Model)
A type of AI model that specializes in understanding and generating text. GPT, Claude, Gemini, and Llama are all LLMs.
Try an LLM in the Playground
Token
The small chunks an AI model breaks your text into before processing it. A token is roughly 3/4 of a word. "Hello, how are you?" is about 6 tokens.
Prompt
The message or instruction you send to an AI model. It's simply what you type in to get a response.
Send your first prompt
Completion
The response an AI model generates after reading your prompt. Also called the output or reply.
Chat Completions
A conversation-style API where you send a list of messages (user, assistant, system) and get a reply back. This is how most AI apps work today.
See the API reference
System Message
A special instruction you give the AI at the start of a conversation to set its behavior — like "You are a helpful coding assistant." The user doesn't see this.
Inference
When an AI model processes your prompt and generates a response. Every time you send a message to an AI, that's one inference.

Context & Limits

Context Window
The maximum amount of text an AI model can look at in a single conversation. A 128K context window means ~128,000 tokens — roughly a 300-page book.
Compare context windows
Max Output Tokens
The longest reply a model can generate in one go. If a model has a 4K max output, its response can be about 3,000 words at most.
Input Tokens
The tokens in your prompt — what you send to the model. You're charged for these.
See pricing per model
Output Tokens
The tokens in the model's reply — what it sends back. These usually cost more than input tokens.
Rate Limit
A cap on how many requests you can send in a given time period (e.g. 60 requests per minute). This prevents any single user from overloading the system.

Model Capabilities

Vision
The ability of a model to understand images, not just text. A vision model can describe a photo, read a screenshot, or analyze a chart.
Try a vision model
Thinking / Reasoning
When a model "thinks step by step" before answering, often producing better results for math, logic, and complex questions. The thinking process may be visible or hidden.
Try a reasoning model
Function Calling / Tools
The ability for an AI model to request that your app run a specific action — like searching a database, checking the weather, or calling another API — and then use the result in its response.
Browse models with tools
Streaming
Getting the AI's response word-by-word as it's being generated, instead of waiting for the entire reply. This makes responses feel faster.
See it live in the Playground
Image Generation
Models that can create images from a text description. You describe what you want and the model draws it.
Try image generation
Video Generation
Models that can create short video clips from a text description.
Try video generation
Embeddings
A way to turn text into a list of numbers that captures its meaning. Similar texts get similar numbers. Used for search, recommendations, and finding related content.
OCR (Optical Character Recognition)
Extracting text from images or PDFs. Antbase offers an OCR bridge that reads documents and returns the text as markdown.
Try OCR in the Playground

Antbase & ANT

ANT
Antbase's intelligent routing engine. It picks the best AI model for each request based on your needs — quality, speed, cost, or capabilities.
Learn how ANT works
Provider
A company that runs AI models and offers them through an API. OpenAI, Anthropic, Google, Groq, and Mistral are all providers. Antbase connects to 30+ of them.
See all providers
Routing
The process of choosing which provider and model should handle your request. ANT does this automatically based on your preferences.
See how routing works
Failover
When a provider goes down or returns an error, ANT automatically retries your request with a different provider so you get a response without doing anything.
Virtual Model
A model name that starts with "ant:" (like ant:auto or ant:fast). It's not a real model — it tells ANT what you care about, and ANT picks the best real model for you.
Explore virtual models
Nest
A custom group of models you create. Use it as "nest:my-name" in API requests, and ANT will pick the best model from that group. Great for limiting which models your app can use.
Learn about nests
Model Lock
Forces all your requests to go to one specific model, bypassing ANT's routing. Useful for testing or when you need consistent behavior.
Set up a model lock
Quality Score
A 0-100 rating Antbase assigns to each model based on benchmark performance, real-world success rates, and community rankings. Higher is better.
See top-ranked models
Free Pool
A collection of models from providers that offer free API access. ANT can route to these when a free model is good enough for the task, saving you money.
Try free models
Model Catalog
The full list of 16,000+ AI models available through Antbase, with their pricing, capabilities, and quality scores.
Browse the catalog
Catalog Sync
The process of pulling the latest model list from each provider and updating the Antbase catalog. This happens automatically so new models appear quickly.

Pricing & Billing

Credits
Prepaid balance on your Antbase account. $1 in credits = $1 worth of AI usage at exact provider rates. No markup on tokens.
Top up credits
Platform Fee
A fee that may be added when you buy credits, shown at checkout. None is charged at present: $100 buys $100 of credit, and there is no markup on actual usage either.
See full pricing details
BYOK (Bring Your Own Keys)
Use your existing API keys from providers like OpenAI or Anthropic. ANT routes through your keys, so you pay the provider directly instead of using Antbase credits.
Set up your keys
Prompt Optimization
ANT can automatically shorten your prompts before sending them to the AI, reducing token usage by 30-50% without changing the meaning. This saves money.
Auto-Downroute
When ANT detects that a cheaper model can handle your request just as well, it routes to the cheaper option automatically. You save money without losing quality.
Input Price / Output Price
What a model charges per 1,000 tokens of input (your prompt) and output (its reply). Output tokens usually cost 2-5x more than input tokens.
Compare model pricing

API & Integration

API (Application Programming Interface)
A way for your code to talk to another service over the internet. You send a structured request and get a structured response back. No UI needed.
Read the API reference
API Key
A secret password that identifies your account when making API requests. Keep it private — anyone with your key can use your account.
Manage your API keys
OpenAI-Compatible
Antbase uses the same request format as OpenAI's API, so any app or tool built for OpenAI works with Antbase by just changing the URL and key.
Quick start guide
Base URL
The web address your app sends API requests to. For Antbase it's https://antbase.ai/v1 — just swap this in where you'd normally put OpenAI's URL.
Get started in 2 minutes
SDK (Software Development Kit)
A library that makes it easier to use an API from your code. Instead of writing raw HTTP requests, you call simple functions like client.chat.completions.create().
See integration guides
Endpoint
A specific URL path that does one thing. For example, /v1/chat/completions is the endpoint for sending messages and getting replies.
See all endpoints
Webhook
A URL that Antbase calls when something happens (like a request finishing). It lets your app react to events without constantly checking.
Workspace
Your team's account on Antbase. A workspace has its own API keys, nests, model settings, and billing. One person can belong to multiple workspaces.
Go to your dashboard

Advanced Concepts

Circuit Breaker
A safety mechanism that temporarily stops sending requests to a provider that keeps failing. Once the provider recovers, traffic resumes automatically.
Orchestrator
The brain of ANT. It receives your request, classifies what you need, picks the best model, handles failures, and returns the response.
See how it works
Classification
The step where ANT analyzes your prompt to understand what it needs — vision? reasoning? a long context window? This helps pick the right model.
Reasoning Effort
How hard you want the model to think. "low" is fast but shallow, "high" is slow but thorough. Useful for balancing speed vs. accuracy.
Try reasoning models
Batch Processing
Sending many requests at once instead of one at a time. Useful for processing large datasets — often cheaper and doesn't count against rate limits.
Latency
How long it takes from sending your request to getting the first word back. Lower latency = faster responses.
See fastest models
Throughput
How many requests a system can handle per second. Higher throughput means more users can be served at the same time.
Open Source Model
An AI model whose code and weights are publicly available. Anyone can run it on their own servers. Llama, Gemma, Qwen, and Mistral are popular open source models.
Browse free open source models
Frontier Model
The most capable AI models available — the cutting edge. GPT-5, Claude Opus, and Gemini Ultra are frontier models. They're powerful but usually the most expensive.
See top-ranked models

Still have questions? Check the docs or reach out.

Read the DocsContact Us