Jurassic-1
AI21 Labs' first widely released language models, in two sizes
Jurassic-1 is a family of language models from AI21 Labs, offered in two sizes, with the Jumbo version at 178 billion parameters. It was released in open beta through the AI21 Studio developer platform, and the company says it was the largest language model then available to developers for general use.
- Jumbo, the larger Jurassic-1 model, has 178 billion parameters
- Two model sizes were released
- Vocabulary of 250,000 tokens, including multi-word tokens
- Custom versions could be trained from 50 to 100 examples
- First AI21 Labs model generation released for wide use
What it is
Jurassic-1 is a set of language models built by AI21 Labs and made available to developers through AI21 Studio. The family comes in two sizes, and the larger one, called Jumbo, has 178 billion parameters. AI21 Labs says this was the biggest language model released at that time for general use by developers. The models generate text that reads like human writing and can also handle tasks such as answering questions and sorting text into categories. It is the first generation of models the company put out for wide use, following research work it has done on language models since 2017.
How it came about
AI21 Labs says it has been working on language models since 2017, a period it jokingly calls its Mesozoic Era. Jurassic-1 grew out of that research and became the first model generation the company released for broad use. It launched together with AI21 Studio, a developer platform that was in open beta at the time, letting anyone sign up and start querying the models straight away through an API and a web environment. The company also published a white paper with technical detail and an evaluation of the models.
Inside the model
The most distinctive design choice is the vocabulary. Jurassic-1 uses a 250,000-token vocabulary, which AI21 Labs says is at least five times larger than most existing ones. It is also described as the first to include multi-word tokens, covering expressions, phrases and named entities. Because whole phrases can be single tokens, less text is needed to represent the same content, which the company says cuts computation and lowers latency. Beyond that, the announcement gives no detail on layers, attention, training data or compute.
What it can do
AI21 Labs describes the models as versatile. They can produce human-like writing and take on harder jobs such as question answering and text classification. The company points to a white paper for a fuller evaluation, but the announcement itself reports no benchmark scores. It also says developers can train custom versions of the models on as few as 50 to 100 examples, with the tuned model served for that customer's own use.
How to use it
Jurassic-1 was offered through AI21 Studio, which was in open beta. Anyone could sign up and query the models through an API or an interactive web environment. Developers could also train custom versions of the models inside the platform, and those were served there for their exclusive use. No weight downloads or licence terms are described.
Why it matters
AI21 Labs says it used AI21 Studio internally to power its own products and that this sped up its development work. The platform was opened so that developers without prior experience could build text applications with the models. Jurassic-1 is presented as the first generation of AI21 Labs models made widely available, coming after several years of the company's own research.
Questions people ask about Jurassic-1
What is Jurassic-1?
It is a family of language models from AI21 Labs, released in two sizes and offered to developers through the AI21 Studio platform. The models generate text and can handle tasks such as question answering and text classification. It was the first model generation the company released for widespread use.
How large is it?
The bigger version, called Jumbo, has 178 billion parameters. AI21 Labs said at launch that this was the largest language model released for general use by developers. A second, smaller size was also part of the family, but its size is not given here.
What is special about its vocabulary?
Jurassic-1 uses a 250,000-token vocabulary, which AI21 Labs says is at least five times larger than most others. It is described as the first to include multi-word tokens such as phrases, expressions and named entities, so less text is needed to represent the same content, lowering computation and latency.
How do developers use it?
They access it through AI21 Studio, which was in open beta, using an API or a web environment. The platform also lets them train custom versions of the models with as few as 50 to 100 training examples, and those tuned models are served there for their own use.
Good reading and sources
There is no English Wikipedia article on Jurassic-1 yet, so the facts on this page come from the sources listed above: ai21.com/blog/announcing-ai21-studio-and-jurassic-1. Some are the subject's own pages, so its claims about itself are reported as its claims. EveryLLM wrote the text in its own words, with AI assistance, and compares it with the sources to make sure no passage is copied.
It may contain mistakes, and figures change over time. Nothing here is investment or buying advice. If something is wrong, tell us. Last refreshed 2026-11-14.