Developer AI APIs and Open-Source Models Directory

Building with AI starts with two choices: which API to call, and whether to run an open model yourself. This page covers both. First, 21 developer APIs we checked in November 2026, each with a link to its official docs. Then a live list of new open models from well-known publishers, with their licenses. We write the notes ourselves and show no prices, since they change often. To compare context windows and token prices, use the AI models comparison. New to the terms? See the AI glossary or the MCP servers directory.

AI APIs for developers (21)

Language model APIs (8)

Official APIs from the labs that build the models.

  • API

    OpenAI API

    Chat, image, speech and embedding models through one API, with the widest tutorial and library support.

    Good for: Most first projects and production apps.

  • API

    Anthropic API

    Access to the Claude models, known for long documents, careful writing and coding help.

    Good for: Writing, analysis and coding tools.

  • API

    Google Gemini API

    Gemini models that read text, images, audio and video, with a long context window.

    Good for: Apps that mix text with images or video.

  • API

    Mistral API

    Open and hosted models from a European lab, including small fast ones.

    Good for: Teams that want open-weight options.

  • API

    Cohere API

    Models focused on business search, retrieval and text embeddings.

    Good for: Search and document question answering.

  • API

    DeepSeek API

    Hosted DeepSeek chat and reasoning models behind an API that follows the OpenAI format.

    Good for: Developers who want a drop-in alternative.

  • API

    xAI API

    Grok models through an API.

    Good for: Apps built around Grok.

  • API

    Perplexity API

    Search-grounded answers with sources returned in the response.

    Good for: Research and answer tools that need citations.

Hosting and routing (6)

Run open models or switch between many models with one key.

  • API

    Groq

    Very fast hosted inference for open models.

    Good for: Chat apps where speed matters.

  • API

    Together AI

    Hosted open models, fine-tuning and image models in one place.

    Good for: Running open models without servers.

  • API

    Fireworks AI

    Hosted inference for open models, with tools for speed and tuning.

    Good for: Production apps on open models.

  • API

    OpenRouter

    One API key for hundreds of models from many labs, with a shared format.

    Good for: Switching or comparing models quickly.

  • API

    Replicate

    Run community image, video, audio and language models by API, paying for what you run.

    Good for: Trying image and video models.

  • API

    Hugging Face Inference

    One interface to run many open models through partner providers.

    Good for: Anyone already on Hugging Face.

Speech, voice and images (4)

APIs for audio and image work.

  • API

    ElevenLabs API

    Text to speech, voice and audio tools by API.

    Good for: Voice apps and narration.

  • API

    Stability AI API

    Image generation and editing models by API.

    Good for: Image tools and product mockups.

  • API

    AssemblyAI

    Speech to text with summaries and speaker labels.

    Good for: Transcribing calls and videos.

  • API

    Deepgram

    Fast speech to text and voice agents by API.

    Good for: Live captions and voice bots.

Building blocks (3)

Databases and toolkits that sit around the model.

  • API

    Pinecone

    A vector database for storing and searching embeddings.

    Good for: Retrieval and memory for chat apps.

  • API

    Vercel AI SDK

    An open-source toolkit for building streaming AI chat in web apps, working with many providers.

    Good for: Next.js and web developers.

  • API

    LangChain

    An open-source framework for chaining model calls, tools and retrieval.

    Good for: Building agents and workflows.

New open-source models

The open model list loads below from Hugging Face's public list.

How we pick

  • APIs: an official provider or a well-known platform, a public developer docs page that opened when we checked, and a clear purpose.
  • Open models: only from a fixed list of well-known publishers, taken from the Hugging Face public model list. We show models under permissive licenses (such as Apache 2.0 and MIT) and the Llama community license, and skip gated models and community re-uploads.
  • No prices, download counts or rankings, because they go stale quickly.

API or open model?

Start with an API

It is the fastest way to a working prototype. You send text, you get an answer, and the provider runs the hardware.

Choose an open model when

You need to keep data on your own machines, want to tune a model, or want to control cost at high volume. Check the license first: some open models limit commercial use.

Hosting services sit in between

They run open models for you and sell them by API, so you can switch later without changing your code much.

Questions developers ask

Is an open-weight model the same as open source?

Not always. Open weights means you can download the model. Open source also needs a license that lets you use, change and share it. The license label on each card helps.

Which API is cheapest?

It depends on the model and your usage. Use the models comparison for current token prices, and test with your own prompts.

Can I use these models commercially?

Apache 2.0 and MIT allow it with few conditions. The Llama license has its own terms. Read the license on the model page before you ship.

How often does the open model list update?

About once an hour.

Why is a model I know missing?

We only show models from our list of publishers with a permissive or Llama license, and skip gated or re-uploaded versions.