AI21 Labs
AI21 Labs, founded in 2017 in Tel Aviv by Yoav Shoham and Amnon Shashua, builds Jamba and Wordtune and raised $636M at $1.4B valuation (May 2025).
Founded: 2017 · HQ: Tel Aviv, Israel · Team: 250+ · CEO: Yoav Shoham · Funding: $200+ million · Valuation: $2+ billion (2023)
About AI21 Labs
AI21 Labs was founded in November 2017 in Tel Aviv, Israel by Yoav Shoham, Ori Goshen, and Amnon Shashua. Yoav Shoham is a Professor Emeritus at Stanford University who previously worked as a Principal Scientist at Google, with research expertise spanning AI, game theory, and multi-agent systems. Amnon Shashua is one of the most accomplished technologists in Israel's history: he is the founder of Mobileye, the advanced driver assistance system company that Intel acquired for $15.3 billion in 2017, and a professor of computer science at the Hebrew University of Jerusalem with foundational work in computer vision. Ori Goshen co-leads as CEO alongside Shoham. The combination of Shoham's language AI expertise, Shashua's track record in building billion-dollar technology companies, and Israel's deep talent pool in mathematics and computer science created the foundation for AI21 Labs. The company's product portfolio spans two distinct categories: foundational language models for enterprise API access and consumer-facing writing tools. The Jurassic model family (including the J2 and L2 variants) were among the first large language models commercially available via API and helped establish AI21 Labs as one of the earliest enterprise AI providers globally. These models were used by thousands of developers through the AI21 API for tasks including text generation, summarization, paraphrasing, and question answering. Jamba is AI21 Labs' most technically distinctive offering. It uses a hybrid Mixture-of-Experts (MoE) architecture that combines the standard Transformer attention mechanism with Mamba, a state space model (SSM) that processes sequences with linear rather than quadratic computational complexity. This hybrid design is particularly well-suited for long-context tasks because the Mamba component efficiently processes long sequences without the quadratic memory and compute costs of pure attention. Jamba supports efficient long-context processing and targets enterprise workloads that require ingesting large documents, codebases, or conversation histories. Wordtune is AI21 Labs' consumer and prosumer writing assistant product, designed to help writers rephrase, summarize, and improve text. It is available as a browser extension, a web application, and integrations with Google Docs and Microsoft Word. Wordtune has attracted millions of users globally and serves as both a standalone product and a demonstration of AI21 Labs' language understanding capabilities. The product competes with Grammarly and similar writing assistants but distinguishes itself through AI-generated rephrasing suggestions that preserve the writer's intent while improving clarity and style. AI21 Labs has raised $636 million in total across 7 rounds as of May 2026. The Series D round of $300 million closed in May 2025 at a $1.4 billion post-money valuation, led by Google and Nvidia. Previous investors include Khosla Ventures, Sequoia Capital, SoftBank, and LG Technology Ventures. The Google and Nvidia co-investment in the Series D is particularly significant because it reflects strategic interest from both a major cloud provider and the dominant GPU manufacturer in AI21 Labs' architectural approach. Revenue crossed $100 million in annual recurring revenue by early 2026. The company is headquartered in Tel Aviv with additional offices in San Francisco, New York, and Boston. It employs between 300 and 500 people, a relatively lean headcount for the volume of research output the team produces. Tel Aviv's AI research community, which benefits from close ties to the Israeli Defense Forces' technology units, the Technion, and the Hebrew University, provides access to a talent pool with deep expertise in applied mathematics and systems engineering. AI21 Labs' research agenda focuses on efficient architectures for enterprise NLP, with particular emphasis on reducing inference costs for long-context workloads. The Jamba architecture's MoE component activates only a fraction of the model's total parameters for each inference step, which reduces per-token compute cost even as the total parameter count grows. This efficiency orientation is a direct response to enterprise procurement patterns, where total cost of ownership is often as important as raw capability when evaluating AI infrastructure vendors. Compliance certifications include SOC 2 Type II, ISO 27001, and GDPR compliance for EU operations. The company's compliance posture supports enterprise and regulated industry sales in North America and Europe. AI21 Labs has positioned itself as a neutral third-party AI provider for enterprises that want frontier model capabilities without committing exclusively to Amazon, Google, or Microsoft's integrated AI ecosystems. AI21 Labs' competitive positioning in the market sits between the pure research labs (Anthropic, DeepMind) and the infrastructure providers (Cohere, Together AI). It offers both a deep research agenda with novel architectural contributions like Jamba and a practical consumer product in Wordtune that demonstrates real-world language understanding. The company's Israeli origin also gives it credibility with enterprise customers in Israel, Eastern Europe, and the broader Middle East and North Africa market, where it has established early commercial relationships. Looking ahead, AI21 Labs' strategy centers on extending the Jamba architecture to support even longer contexts and more efficient inference, expanding Wordtune's enterprise capabilities to compete more directly with Grammarly Business, and deepening the relationships with Google and Nvidia established through the Series D to access more compute at preferential terms. The $1.4 billion valuation and $100 million in ARR suggest the company is approaching the threshold needed for a credible public market debut, though no specific IPO timeline has been announced.
Mission
To develop practical AI technologies and applications that solve real business problems while maintaining ethical standards and responsible deployment practices.
Products
- Jurassic API
- Jurassic-2 Models
- AI21 Writing Assistant
- Domain-Specific Applications
Compliance
SOC 2 Type II
Links
Website · GitHub · Twitter · LinkedIn · Blog · Docs
Frequently Asked Questions
Who founded AI21 Labs and what is their background?
AI21 Labs was founded in November 2017 in Tel Aviv, Israel by three individuals with distinct and complementary backgrounds. Yoav Shoham is a Professor Emeritus at Stanford University who previously worked as a Principal Scientist at Google, with research expertise in AI, game theory, and multi-agent systems. Amnon Shashua is the founder of Mobileye, the advanced driver assistance system company that Intel acquired for $15.3 billion in 2017; he is also a professor of computer science at the Hebrew University of Jerusalem and a pioneer in computer vision research. Ori Goshen co-leads the company as co-CEO alongside Yoav Shoham. Shashua's track record of building and scaling a multi-billion-dollar technology company in Israel was a significant factor in AI21 Labs attracting early-stage capital and talent. The founding team chose Tel Aviv because of Israel's deep talent pool in applied mathematics, computer science, and systems engineering. The combination of Shoham's language AI academic expertise and Shashua's industry-building experience created an unusually strong founding team for an AI startup.
What is the Jamba model and what makes it architecturally distinctive?
Jamba is AI21 Labs' flagship model architecture that uses a hybrid design combining the standard Transformer attention mechanism with Mamba, a state space model (SSM) that processes sequences with linear rather than quadratic computational complexity. Pure Transformer models have quadratic memory and compute scaling with sequence length, which makes very long contexts expensive; Mamba's linear scaling addresses this bottleneck. Jamba also uses a Mixture-of-Experts (MoE) design, which means only a fraction of the model's total parameters are activated for each inference step, reducing per-token compute cost relative to dense models with the same parameter count. The architecture supports efficient long-context processing and is particularly well-suited for enterprise workloads involving large documents, codebases, or extended conversation histories. AI21 Labs released Jamba under a commercial license for enterprise API access. The hybrid Transformer-Mamba design is one of the first commercial implementations of an SSM-augmented architecture to be deployed at significant scale, making it a technically important development in the broader trend toward more efficient long-context models.
What is Wordtune and how does it differ from other writing assistants?
Wordtune is AI21 Labs' AI writing assistant, available as a browser extension, web application, and integrations with Google Docs and Microsoft Word. It is designed to help writers rephrase, improve, and summarize text while preserving the writer's original intent and voice. Wordtune has attracted millions of users globally across individual, student, and professional use cases. The product competes with Grammarly and similar writing assistants but differentiates itself through AI-generated rephrasing suggestions that go beyond grammar correction to offer alternative phrasings that change tone, formality, or conciseness according to the writer's preference. Wordtune's AI capabilities draw on AI21 Labs' core language model research, giving it stronger semantic understanding of what a sentence is trying to communicate compared to rule-based grammar tools. A Wordtune for Business product serves enterprise teams with additional administrative controls, analytics, and brand voice customization. The product is available in free and paid tiers, with the premium plan priced for individual power users and a separate enterprise pricing track.
How much funding has AI21 Labs raised?
AI21 Labs has raised $636 million in total across 7 funding rounds as of May 2026. The most recent round was a $300 million Series D closed in May 2025 at a $1.4 billion post-money valuation, led by Google and Nvidia. Earlier rounds include a $155 million Series C and a $64 million Series B. Investors across all rounds include Khosla Ventures, Sequoia Capital, SoftBank, and LG Technology Ventures, in addition to the strategic investors Google and Nvidia who joined in the Series D. The co-investment by Google and Nvidia in the Series D reflects strategic interest in AI21 Labs' architectural approach, particularly the Jamba hybrid Transformer-Mamba design, which aligns with both companies' interests in efficient long-context inference. Revenue crossed $100 million in annual recurring revenue by early 2026. No IPO timeline has been announced, but the $1.4 billion valuation and $100 million ARR put the company in range of a credible public market debut in the next several years. The company's Israeli roots mean it would likely consider a Tel Aviv Stock Exchange or NASDAQ listing.
What is AI21 Labs' connection to Mobileye and Amnon Shashua?
Amnon Shashua is a co-founder of AI21 Labs and also the founder of Mobileye, one of the most successful technology exits in Israeli history. Mobileye pioneered vision-based driver assistance systems, including lane departure warning, automatic emergency braking, and other ADAS features, using camera-based sensing rather than radar or lidar. Intel acquired Mobileye for $15.3 billion in 2017, and Mobileye subsequently went public on NASDAQ in October 2022 at a $17 billion valuation, making it one of the largest Israeli technology IPOs. Shashua is also a professor of computer science at the Hebrew University of Jerusalem with foundational research in computer vision and machine learning. His involvement in AI21 Labs brings both the operational expertise of building and scaling a multi-billion-dollar technology company and the academic credibility of a leading researcher in the field. Shashua's dual role as a serial entrepreneur and active professor is consistent with the broader Israeli technology culture of combining academic research with applied commercialization. His track record has been instrumental in AI21 Labs attracting investors including SoftBank, Sequoia Capital, and Google.
What compliance certifications does AI21 Labs hold?
AI21 Labs holds SOC 2 Type II certification, which provides third-party assurance of security, availability, processing integrity, confidentiality, and privacy controls. The company is certified to ISO 27001 for information security management. AI21 Labs complies with GDPR for EU operations through data processing agreements and appropriate data handling practices. These compliance credentials support AI21 Labs' enterprise sales in North America, Europe, and Israel's regulated industry sectors including financial services, healthcare, and government. The company's Tel Aviv headquarters and additional offices in San Francisco, New York, and Boston allow it to serve customers across US, European, and Israeli regulatory jurisdictions. AI21 Labs positions itself as a neutral enterprise AI provider that can serve customers who want frontier model capabilities without committing exclusively to one of the large cloud providers. Additional enterprise security features including audit logging, access controls, and data residency options are available to enterprise API customers.
What markets does AI21 Labs focus on?
AI21 Labs focuses on enterprise NLP and consumer writing assistance across North America, Europe, and Israel. The enterprise API business targets organizations that need large language model capabilities for text generation, summarization, paraphrasing, and question answering, particularly in industries with long-document workflows including legal, financial services, healthcare, and media. Wordtune serves a global user base of millions of individual users, students, and professionals who want AI-assisted writing improvement. AI21 Labs has particular commercial strength in Israel, Eastern Europe, and the broader Middle East and North Africa region, where its Israeli origin provides cultural and relationship advantages that US-headquartered AI companies do not have. The Series D investment from Google and Nvidia suggests potential for deeper integration with Google Cloud and Nvidia's AI infrastructure partner ecosystem, which would expand AI21 Labs' reach in the enterprise market. The company's lean headcount of 300-500 employees reflects a deliberate strategy of maintaining high output per researcher and focusing R&D investment on architecture innovation rather than broad horizontal product expansion.
How does AI21 Labs' Jurassic model family compare to newer offerings?
The Jurassic model family (including J2 and L2 variants) was AI21 Labs' original commercial offering and represented one of the earliest third-party alternatives to OpenAI's GPT-3 when it launched in 2021. The Jurassic models were trained on large text corpora and offered competitive performance on standard language benchmarks for their size class. They were made available through the AI21 API for tasks including text generation, question answering, and summarization, and attracted thousands of developers during the 2021-2023 period. However, as the field advanced rapidly through 2023 and 2024, AI21 Labs pivoted significant research resources toward the Jamba architecture, which offers a fundamentally different approach to long-context efficiency through its hybrid Transformer-Mamba design. Jamba represents the company's current technical frontier, while the Jurassic family continues to serve existing customers. AI21 Labs has been transparent about the architectural advantages of Jamba over the Jurassic family for long-context enterprise workloads. The company expects Jamba to be its primary enterprise offering going forward, with newer model versions extending context length and reducing inference cost.