Start with a free trial, no credit card required, then pay based on usage.

Starter Plan

Access all features through usage-based pricing,
with a minimum subscription fee of $29 per month.

Access all features through usage-based pricing, with a minimum subscription fee of $29 per month.

  • Usage-based pricing

  • Foundation model APIs & SDK

  • Custom model training

  • Task-specific models

  • Unlimited seats

Free Trial

  • $90 credits for 3 months

  • No credit card needed

Start Now For FREE

Custom Plan

For developers who are looking to scale, require custom implementation or dedicated support.

  • Everything in the Starter Plan

  • Volume discounts

  • Premium API rate limits

  • Private cloud hosting

  • Priority support

  • Dedicated account manager

  • Expert AI consultancy

Foundation Models

Instruct and Custom models

Jurassic foundation models are trained on a massive corpus of text, making them a powerful basis for a wide range of natural language processing applications, capable of understanding & composing human-like text.

Ranked best-in-class by Stanford University's HELM index.

Train custom models with exceptional flexibility.

Use your fine-tuned models at no extra cost.


Highest quality

$0.015 / 1k tokens


Train a custom model


/MB /epoch


Optimal balance of quality, speed & cost

$0.01 / 1k tokens


Train a custom model


/MB /epoch


Fastest and most cost-effective

$0.003 / 1k tokens


Train a custom model


/MB /epoch

What is a Token?

Tokens are used in language models to break down text into small pieces. With AI21 Studio's models, an average token corresponds to 1 word or 6 characters of English text. Comparatively, this is up to 30% more text per token than other providers, saving you 30% of the cost.

Task-Specific Models

Specialized APIs tailored for reading and writing

Task-Specific Models offer out-of-the-box reading and writing capabilities, enabling developers to build AI-powered apps without any NLP expertise or additional model training. Powered by highly optimized models, the models deliver industry-leading results and strong, reliable performance at low latency.

Outperforms main market solutions in quality and latency.

No prompt-engineering or fine-tuning needed.

Plug-and-play NLP, accessible to any developer.

Wordtune API


Takes a piece of text and returns a list of paraphrases that convey the same meaning using different words.

$0.001 / API request

Text Improvements

Provides rewrite recommendations for improving the fluency, clarity and style of the input text.

$0.0005 / API request

Grammatical Error Corrections

Detects grammar, spelling, punctuation, and word choice errors in text, and suggests how to correct them.

$0.0005 / API request


Takes a document (raw text or URL) and returns a summary of the text.

$0.005 / API request

Text Segmentation

Takes a document (raw text or URL) and returns segmentation into paragraphs by topic.

$0.001 / API request

Summarize by Segment

Takes a document (raw or URL), segments it into paragraphs by topic and returns a summary of each paragraph.

$0.0005 / API request

Contextual Answers API

Contextual Answers


Receives documents and returns an answer to a question based entirely on the provided context of the documents.

$0.005 / API request



Stores uploaded documents on a vector database with their embeddings.


Semantic Search


Receives documents and returns the most relevant text segments and file reference in response to a query.

$0.004 / API request

Frequently Asked Questions

Can I still use Jurassic-1 models?

Jurassic-1 models remain available until June 21st, 2023, after which they will be deprecated in favor of Jurassic-2.

What is the difference between training and usage rates for fine-tuned models?

Custom models cost the same to use as pre-trained foundation models. However, training a custom model requires a one-time fee. The cost is dependent on the selected model, and is calculated by the size of the dataset, measured in megabytes, multiplied by training duration, measured in epochs (each epoch is one complete run of the entire training dataset).

What is an epoch, and how many epochs are required to train a custom model?

When training a custom model, an epoch is defined as one pass of the entire training dataset. Typically, models are trained for 5-20 epochs, though finding the optimum number of epochs may require experimentation.

How can I track my usage and spending?

To track usage, log in and navigate to 'Account,' then click 'Model Usage'. This page shows a detailed view of your organization's usage for the current billing period. To see your future expected payment and previously paid invoices, navigate to 'Account', and then click 'Billing & Plans'.

When does the billing cycle begin?

Usage is billed monthly, starting on the date you upgraded to a paid plan. Each month, the billing cycle is the same. On the day you upgrade, your usage for the month will be billed, and then the next month's billing cycle will start on the same date. This will continue on a monthly basis until you switch plans or cancel your subscription.

Who is eligible for the volume discount?

Volume discounts are available to customers with high monthly usage, regardless of whether they are developers, startups, enterprises, or research teams. This type of discount offers a reduced rate for increased usage, when subscribing to a custom plan. Our sales team will be happy to tailor an offer to meet your requirements and volume of business.

Why am I no longer seeing the Jurassic-2 Large, Grande and Jumbo models?

The Jurassic-2 family of models has been renamed to make it easier for you to choose the model that’s right for you. The current lineup of models are:

Jurassic-2 Ultra (formerly Jurassic-2 Jumbo Instruct)

Jurassic-2 Mid (formerly Jurassic-2 Grande Instruct)

Jurassic-2 Light (formerly Jurassic-2 Large with added instruct capabilities)

All three foundation models follow natural-language instructions and allow for both zero-shot and few-shot prompting. For more information, please read here.