Live session Oct 22: AI Search for Clients
Promptwatch Logo

AI API

An AI API is a hosted interface for sending prompts, files and tool definitions to a provider's model, such as GPT-6, Claude or Gemini, and getting output back.
Updated October 15, 2026
AI

Definition

An AI API is a programmatic interface to a model hosted by a provider. An application sends input (text, images, audio, files, tool definitions) and receives generated output, without running the model itself. Most AI features in software, from support chat to search assistants and coding tools, are built this way.

As of October 2026, the main first-party APIs and their current flagship models are:

  • OpenAI: GPT-6 Astra ($10 per million input tokens, $50 per million output), GPT-6.1 Sol ($2 / $10) and GPT-6 Luna ($0.10 / $0.50), through the Responses and Chat Completions APIs
  • Anthropic: Claude Opus 5.5 ($4 / $20), Claude Sonnet 5.5 and Claude Haiku 5.5, plus Claude Fable 5.1 ($10 / $50)
  • Google: Gemini 4 Argon, generally available since October 13, 2026 at an introductory $2 / $10 until November 30, 2026 and $4 / $20 after that, alongside Gemini 3.8 Flash

Cloud platforms resell many of these. Amazon Bedrock, Google Cloud and Microsoft Foundry all carry the Claude 5.5 models, and open-weight models such as Qwen, DeepSeek, Gemma and gpt-oss are served by many inference providers or can be self-hosted.

Beyond text generation, the standard feature set now includes function calling and built-in tools such as web search and code execution, structured outputs, image and document input, discounted batch processing, prompt caching and embeddings. Pricing is per token, with output several times more expensive than input and cached input much cheaper. Context windows of about one million tokens are now normal at the top end: GPT-6 Luna, Claude Opus 5.5 and Gemini 4 Argon all list roughly 1M. Running the model is AI inference, and its cost is what these prices reflect.

Models are replaced quickly, so API users plan for change. Anthropic has scheduled Claude Sonnet 4.5's retirement from its API for November 30, 2026, and OpenAI's documentation says it is winding down its fine-tuning platform and no longer accepts new users. Teams that hard-code one model ID, or depend on a feature only one model has, tend to get caught out.

APIs matter for AI visibility in two ways. Every app built on them is another place where a brand can be mentioned, often with web search grounding switched on, so citations can come from tools you have never heard of. And API output is not the same as what users see in consumer apps: the ChatGPT or Gemini app adds its own search layer, system instructions and interface. Tracking visibility through the API alone can miss what real users get, the trade-off covered in UI vs API tracking and in our post on UI vs API for AI search visibility.

Examples of AI API

  • A legal tech product routes routine clause extraction to GPT-6 Luna at $0.10 per million input tokens and sends only flagged clauses to Claude Opus 5.5 for a second review.
  • A travel startup calls Gemini 4 Argon with Google Search grounding so itinerary answers cite current opening hours instead of relying on training data.
  • An enterprise runs Claude Haiku 5.5 through Amazon Bedrock to keep data, access controls and billing inside its existing AWS account.
  • A GEO team compares a brand's mentions in raw API responses with captured answers from the ChatGPT app for the same prompts and finds the two cite different sources.

Terms related to AI API

AI Inference

AI inference is running a trained model to produce output from new input. Every ChatGPT reply, AI Overview and agent step is inference, usually billed by token.

AI

Tokens

Tokens are the units of text a language model reads and writes, such as word pieces or punctuation. They set context window limits, API prices and rate limits.

AI

Function Calling / Tool Use

Function calling, or tool use, lets an LLM request a defined function, such as a search or API call, with structured arguments that an app then runs.

AI

OpenAI

OpenAI is the AI company behind ChatGPT, Codex and the GPT-6 models. It said in September 2026 that ChatGPT had 1.2 billion weekly users.

AI

Anthropic

Anthropic is the AI company behind Claude and the Model Context Protocol, known for constitutional AI and safety research, and valued at $965B in May 2026.

AI

Google Gemini

Google Gemini is Google's AI model family and assistant app. Gemini models power AI Overviews and AI Mode; Gemini 4 Argon is the newest flagship.

AI

Embeddings

Embeddings are numeric vectors that represent the meaning of text, images or other data, so AI systems can compare content by meaning. They power vector search.

AI

UI vs API Tracking

UI vs API tracking is the choice between collecting AI answers from a chatbot's consumer interface or its API, which return different sources and brands.

Analytics

Context Window

A context window is the maximum number of tokens an LLM can consider at once, covering input and output. Frontier models offered about 1 million tokens in 2026.

AI

Open Source LLMs

Open source LLMs are language models whose weights anyone can download and run. Most are open-weight, not fully open source, and come with varied licenses.

AI

Frequently Asked Questions about AI API

Learn about AI visibility monitoring and how Promptwatch helps your brand succeed in AI search.

Most AI APIs charge per million tokens, with separate rates for input, cached input and output. For the current GPT-6, Claude 5.5 and Gemini 4 models, output costs about five times the input rate. Reasoning models also bill their internal thinking as output tokens, so complex questions cost more. Batch processing and prompt caching cut costs, and some providers charge higher rates for very long prompts.

Be the brand AI recommends

Monitor your brand's visibility across ChatGPT, Claude, Perplexity, and Gemini. Get actionable insights and create content that gets cited by AI search engines.

Promptwatch Dashboard