PlaygroundAPI KeyAccount

Overview

Welcome to AI21 Studio!

AI21 offers affordable, best-in-class large language model (LLM) solutions for many common business tasks, including:

  • Querying document libraries
  • Implementing help chatbots
  • Classifying the sentiment of text snippets
  • Ad copy suggestions

To support these actions, AI21 has developed two different types of models:

  • Foundation models: These are very large, general-purpose LLM models capable of a large variety of tasks, such as question answering, text completion, and chat.
    • Jamba Instruct, our top-notch foundation model, offers exceptional flexibility and efficiency for a wide range of business use cases. Designed to surpass traditional language models, Jamba supports extensive language capabilities, enhancing its applicability for global use.
    • Our older Jurassic-2 series, equipped with high-performance language models, encompasses three variants - Ultra, Mid, and Light. These models are distinguished by their unwavering delivery of superior performance and broad flexibility, excelling in diverse tasks.
  • Task-Specific Models: These are models optimized for specific tasks such as question answering limited to one or more documents, document semantic search, and grammatical suggestions and improvements. Task-Specific Models can provide better output and faster response for specific use cases than foundation models.

Our models cannot be fine-tuned, but you can include an large amount of context (256K) in your prompts to the Jamba foundation model, or use our RAG Engine to include a personal library of documents as the basis for your queries.

Accessing our models

You can access our tools and models through several different mechanisms:

  • Python SDK: We provide a Python SDK to simplify access to all our models and tools from your Python code. The SDK provides code completion tips, documentation, support for synchronous and asynchronous calls, and much more.
  • Cloud platform implementations: We provide model access from AWS SageMaker and Bedrock. Models are accessed through platform-specific wrappers provided by your cloud platform, or by using the Python SDK in your code on that platform.
  • An interactive playground: Try out our various models and tools without any code in our browser-based web app. These are good for experimenting with the models and viewing output in a small scale, or for small scale usage. You must sign up with AI21 in order to use the web app.
  • REST API: Under the hood, our SDK, playground, and cloud platform implementations access our models through our public REST API. If you prefer making API calls yourself rather than using one of these intermediaries, you can call these endpoints directly in your code.
  • Other third-party services: Our models are also available on other third-party systems such as LangChain and LlamaIndex. Check your toolchain to see if AI21 is available.

Who are these docs for?

Our documentation explains our models and tools, the basics of LLMs, and how to implement the most common use cases, as well as API, SDK, and cloud usage details. You do not need a background in AI or LLMs to use our products.

If you are not a programmer, you can still get the gist of our product in the Guides section of our docs. If you want to try out our models without any code, you can use our product playground. If you want to access our products through any other method, you should have programming experience.

Next steps

Updated 13 days ago