AI21 Labs Ltd. is an Israeli company specializing in generative artificial intelligence and natural language processing. Based in Tel Aviv, it combines research into language models with the development of products for individuals and businesses. Its activities thus span two complementary uses: helping users write directly and providing developers with the components they need to build their own applications.
A company rooted in artificial intelligence research
AI21 Labs was founded in 2017 by Amnon Shashua, Yoav Shoham and Ori Goshen. The first two are recognized researchers in artificial intelligence; Amnon Shashua is also a co-founder of Mobileye, a company specializing in driver-assistance technologies. This scientific background shapes AI21’s positioning: it develops its own models rather than limiting itself to an interface built on those of other providers.
The company became known to the general public through Wordtune, launched in 2020. This assistant offers features such as rephrasing texts and adjusting their tone. AI21 subsequently expanded its offering with the Jurassic model family, then introduced Jamba in 2024, in a market now structured around several providers of general-purpose models.
From writing tools to enterprise models
Wordtune serves as a showcase for AI21’s applications. The service supports writing, rephrasing and content summarization. It illustrates an approach focused on assisting users: suggesting different phrasings, clarifying a passage or making it easier to work with a document, rather than systematically replacing the entire writing process.
For developers and organizations, AI21 also offers access to its technologies through application programming interfaces and related services, notably through AI21 Studio. Intended uses include text generation, summarization, document analysis and answering questions based on information supplied by a company.
The Jamba family stands out for its hybrid architecture combining Transformer and Mamba components, with the latter approach belonging to the category of state-space models. This technical choice aims in particular to process long text sequences while keeping computing and memory requirements under control. For business customers, however, its value is measured under real-world conditions: response quality, latency, operating costs and the ability to work with large documents.
What next?
AI21’s trajectory will depend on its ability to turn its research choices into lasting operational benefits. When competing with providers of proprietary models as well as open-weight models, differentiation is no longer based solely on general performance. Integration with existing systems, data privacy, reliability on specific tasks and cost predictability are becoming decisive factors. For AI21, the challenge is therefore to consolidate its position in professional use cases where language processing delivers measurable gains, without overlooking the need for human oversight and evaluation.