AI21 Labs Jurassic

AI21 Labs, founded in 2017 by Prof.

Updated 12 Jun 2026 Reviewed by 7wData

Visit Website

Profile

AI21 Labs develops enterprise AI systems, from language models to planning-and-orchestration platforms that improve reliability and accuracy of AI agents in production.

AI21 Labs, founded in 2017 by Prof. Yoav Shoham (Stanford computer scientist, formerly Google Principal Scientist), Ori Goshen, and Prof. Amnon Shashua (co-founder of Mobileye), operates as an AI lab and product company based in Tel Aviv, Israel.

The company raised $208 million in a Series C round in November 2023 at a $1.4 billion valuation, reaching cumulative funding of $336 million at that time. In 2025, it generated approximately $57.8 million in revenue. AI21 initially built its reputation through the Jurassic family of language models—Jurassic-1 (2021) and Jurassic-2—which were deployed across enterprise use cases in finance, healthcare, manufacturing, and defense.

The company also operated Wordtune, a consumer writing assistant, until discontinuing it in April 2025 as part of a strategic pivot. In August 2024, AI21 released Jamba, a hybrid SSM-Transformer architecture model family with 256K-token context windows, available on AWS Bedrock, Google Cloud, and other cloud platforms. However, 2025 marked a dramatic strategic shift.

The company discontinued standalone model sales in December 2025, cutting its workforce from approximately 180 to 70 employees—a 60% reduction—to focus exclusively on Maestro, its AI planning and orchestration system announced in March 2025. Maestro addresses enterprise reliability challenges by combining structured planning, execution, and validation; it improves GPT-4o accuracy from 85% to 91.9% on instruction-following benchmarks. AI21 signed contracts worth tens of millions of dollars for Maestro implementations, including partnerships with Wix (website builder) and Nebius (infrastructure company), which began as acquisition discussions before pivoting to a commercial partnership.

The restructuring reflects AI21's assessment that selling models alone was not sufficiently sustainable, and that enterprises demand specialized agent optimization solutions rather than general-purpose language models. As of late 2025, acquisition discussions with Nvidia were reported at valuations between $2–3 billion, suggesting an exit path above the company's last disclosed $1.4 billion valuation.

Who buys this

Financial services firms optimizing compliance and document analysis workflows
Healthcare and pharmaceutical enterprises building AI-assisted research and customer support
Technology companies (including Wix) integrating AI agents into product platforms
Defense and manufacturing sectors deploying language models for knowledge automation
Research institutions and academic publishers (Clarivate partnership for literature discovery)

Publicly disclosed clients

Clarivate (partnership for conversational academic search via AI21 language models)
Wix (integrating Maestro for AI-powered platform features)
Nebius (licensing Maestro for infrastructure and AI agent operations)
Google Cloud (partnership for deployment of AI21 models on Vertex AI)
Amazon Web Services (Jurassic-2 models available via Bedrock)

Strengths and what to watch

Strengths

Novel hybrid SSM-Transformer architecture (Jamba) delivers 256K context windows and superior latency on long-context tasks vs. competitors like Llama and Mixtral
Maestro orchestration system demonstrates 50% accuracy improvements over standalone LLMs on instruction-following tasks, addressing real enterprise pain point
Deep partnerships with major cloud providers (Google, AWS, Microsoft) and infrastructure leaders (Nebius, Wix) provide distribution and embedded revenue models

Watch for

Dramatic December 2025 restructuring (60% workforce cuts, discontinued model sales) indicates revenue per engineer was unsustainable; execution risk on Maestro pivot is material
Founder Amnon Shashua co-launched separate AI venture (AAI) in late 2023, signaling declining internal confidence; founding team stability is uncertain
Reported Nvidia acquisition talks at $2–3B valuation suggest pressure to exit; integration or collapse of deal would reshape product roadmap and customer relationships

Recent moves

Key Information

Industry: Closed Source Models
Founded: 2017
Headquarters: Tel Aviv, Israel

Frequently Asked Questions

What is AI21 Labs?

AI21 Labs, founded in 2017 by Stanford scientist Yoav Shoham, Ori Goshen, and co-founder of Mobileye Amnon Shashua, develops enterprise AI systems. These include the Jurassic family of language models and Maestro, an AI planning and orchestration platform designed for production reliability.

What does Maestro do?

Maestro is an AI planning and orchestration system that improves agent reliability through structured planning, execution validation, and error recovery. Announced in March 2025, it demonstrates a 50% accuracy improvement over standalone models. Applied to GPT-4o, Maestro boosts instruction-following accuracy from 85% to 91.9%.

What is Jamba?

Jamba is AI21's language model family released in August 2024 featuring a hybrid SSM-Transformer architecture and 256K-token context window. The hybrid design delivers faster inference speeds and superior performance on long-context tasks, outperforming competitors like Llama and Mixtral on efficiency and accuracy metrics.

Why did AI21 Labs restructure in 2025?

In December 2025, AI21 Labs cut its workforce by 60%—from 180 to 70 employees—and discontinued model sales. This restructuring reflects a pivot toward Maestro, AI21's AI orchestration system. The company determined general-purpose model sales were unsustainable; enterprises instead demand specialized agent-optimization solutions that improve reliability.

How much can Maestro improve AI accuracy?

Maestro improves AI agent accuracy through structured planning, execution, and error validation. Demonstrations show a 50% accuracy improvement over standalone language models on instruction-following tasks. When applied to GPT-4o, Maestro boosts accuracy from 85% to 91.9%, directly addressing the enterprise AI reliability challenge customers face.

Which companies use AI21 Labs?

AI21's customers include Wix, which integrates Maestro for AI-powered features, Nebius for infrastructure AI, and Clarivate for conversational academic search. Beyond individual clients, Google Cloud and AWS distribute AI21's models through Vertex AI and Bedrock. Customer sectors span finance, healthcare, defense, and manufacturing.

How AI21 Labs Jurassic compares

Direct head-to-head against 2 competitors. Picked by 7wData.

Positioning: AI21 Labs develops enterprise AI systems, from language models to planning-and-orchestration platforms that improve reliability and accuracy of AI agents in production.
Customer segments: Financial services firms optimizing compliance and document analysis workflows
Strengths: Novel hybrid SSM-Transformer architecture (Jamba) delivers 256K context windows and superior latency on long-context tasks vs. competitors like Llama and Mixtral
Watch for: Dramatic December 2025 restructuring (60% workforce cuts, discontinued model sales) indicates revenue per engineer was unsustainable; execution risk on Maestro pivot is material
Recent moves: Nvidia in advanced talks to acquire AI21 Labs for $2–3 billion, reports indicate

Positioning: Enterprise AI platform for data sovereignty, RAG with citations, and flexible on-premises and cloud deployment.
Customer segments: Finance, retail, and regulated industries requiring air-gapped or sovereign deployment; named customers include RBC, Oracle, and McKinsey.
Strengths: North platform delivers citation-grounded retrieval in a security-first AI workspace deployable in VPC or air-gapped environments.
Watch for: April 2026 Aleph Alpha merger adds integration execution risk; combined entity must prove transatlantic go-to-market coherence before IPO.
Recent moves: April 2026: announced merger with Aleph Alpha, backed by $600M Schwarz Group investment, creating a transatlantic enterprise AI platform.

Positioning: European model and orchestration platform targeting regulated industries in finance, aerospace, and automotive with sovereignty positioning.
Customer segments: Large European enterprises in aerospace, automotive, and financial services; named customers include Airbus, BMW Group, ASML, HSBC, and BNP Paribas.
Strengths: April 2026 Agents API and Workflows launch positions Mistral as an orchestration competitor targeting enterprise RPA and document-processing budgets.
Watch for: 2025 safety audit found Pixtral vision models generate harmful content at rates up to 60x higher than GPT-4o.
Recent moves: September 2025: raised 1.7 billion EUR Series C at 11.7 billion EUR valuation, led by ASML.

Sources

www.ai21.com — Founding team (Amnon Shashua, Ori Goshen, Yoav Shoham), mission statement, current products (Maestro, Jamba)
www.prnewswire.com — Series C funding amount ($208M), date (November 2023), valuation ($1.4B), cumulative funding ($336M)
cloud.google.com — Google Cloud partnership, deployment of Jurassic models, use of GKE, Cloud Storage, BigQuery for AI21 operations
www.calcalistech.com — Wordtune discontinuation (April 2025), revenue ($57.8M in 2024), failed funding attempts, Amnon Shashua's separate AI venture, strategic pivot to Maestro
www.calcalistech.com — December 2025 layoffs (60% reduction, 180→70 employees), discontinuation of standalone model sales, Maestro focus, Nebius partnership, Wix partnership
www.prnewswire.com — Maestro announcement (March 2025), accuracy improvements (GPT-4o 85%→91.9%), SaaS and VPC availability timeline
www.ai21.com — Jamba model architecture, 256K context window, hybrid SSM-Transformer design, enterprise deployment
www.prnewswire.com — Clarivate partnership for conversational academic search, two-week prototype deployment timeline