AI21 Labs generated an estimated $57.8 million in revenue during 2024 while holding just 0.01% of the enterprise LLM market — a gap between technical performance and commercial traction that defines the company’s position heading into 2026. This article compiles the latest AI21 Jurassic statistics covering funding, model benchmarks, market share, and what the shift to Jamba means for the Jurassic model family.
AI21 Jurassic Statistics 2026 – TL;DR
AI21 Labs recorded $57.8 million in 2024 revenue with approximately 278 employees, putting revenue per employee around $208,000. Jurassic-2 Ultra scored an 86.8% win rate on HELM benchmark evaluations. The company raised $636 million in total funding through its May 2025 Series D round. AI21 scored 75 on Stanford’s Foundation Model Transparency Index, well above the industry average of 58. Nvidia entered acquisition talks valued at $2–$3 billion in late 2025, but those negotiations fell through by early 2026 — and Nebius is now reported to be in discussions to acquire the company.
AI21 Jurassic statistics for 2026 show a company at a crossroads. The Jurassic-2 model family, trained on 1.2 trillion disclosed tokens, still handles enterprise workloads through AI21 Studio. But the company’s own focus has moved to its Jamba models, which use a hybrid Mamba-Transformer architecture and offer a 256K token context window. The enterprise generative AI adoption wave continued to grow, but AI21’s slice of it stayed small against competitors like OpenAI, Anthropic, and Google.
How Much Revenue Does AI21 Labs Generate?
AI21 Labs reported $57.8 million in revenue for the 2024 fiscal year. Some unverified estimates place the upper range as high as $111.6 million, though the lower figure is more widely cited. Annualized revenue was reported at roughly $50 million as of late 2025, according to Calcalist’s reporting during the Nvidia acquisition talks.
With 278 employees, the company’s revenue per employee sits around $208,000. That figure reflects the heavy R&D spending typical of foundation model companies at this stage. For context, Anthropic reached an estimated $5 billion in annualized revenue by mid-2025, and Cohere was tracking at $150 million ARR by October 2025.
| Metric | Value |
|---|---|
| 2024 Revenue | $57.8M |
| Annualized Revenue (Late 2025) | ~$50M |
| Employee Count | ~278 |
| Revenue Per Employee | ~$208,000 |
| Enterprise LLM Market Share | 0.01% |
Source: Contrary Research, SiliconANGLE, Calcalist
AI21 Jurassic Funding History
AI21 Labs has raised $636 million across seven funding rounds since 2019. The most recent was a $300 million Series D in May 2025, led by Nvidia and Alphabet. Google and Nvidia also participated in the $155 million Series C round in August 2023, which valued the company at $1.4 billion.
An earlier report suggested the Series D round was not fully completed or officially disclosed, adding some ambiguity to the total capital figure. Regardless, AI21 sits well below the funding levels of OpenAI and Anthropic, both of which have raised tens of billions.
| Round | Date | Amount | Notable Investors |
|---|---|---|---|
| Seed | Jan 2019 | $9.5M | — |
| Series A | Nov 2021 | $25M | Pitango First |
| Series B | Jul 2022 | $64M | Ahren, Walden Catalyst |
| Series C | Aug 2023 | $155M | Google, Nvidia |
| Series D | May 2025 | $300M | Nvidia, Alphabet |
Source: Tracxn, Wikipedia, Contrary Research
How Does Jurassic-2 Perform on Benchmarks?
Jurassic-2 Ultra achieved an 86.8% win rate on internal HELM (Holistic Evaluation of Language Models) evaluations. At the time of its March 2023 release, that placed it among the top-performing LLMs. The mid-sized Jurassic-2 Grande also ranked higher than models up to 30 times its size on HELM, according to AI21’s own testing.
Jurassic-2 was trained on 1.2 trillion tokens drawn from a filtered CommonCrawl dataset, Wikipedia, BookCorpus, and arXiv/StackExchange data. All three variants — Light, Mid, and Ultra — support zero-shot instruction-following. The model family’s training data extended up to mid-2022, making it dated compared to models released in 2024 and 2025.
| Model | Use Case | HELM Win Rate |
|---|---|---|
| Jurassic-2 Ultra | Complex generation, multilingual | 86.8% |
| Jurassic-2 Mid | Balanced quality and cost | Not disclosed |
| Jurassic-2 Light | High-volume, low-cost tasks | Not disclosed |
Source: AI21 Labs, Stanford CRFM
AI21 Jurassic Transparency Score
AI21 scored 75 on Stanford’s Foundation Model Transparency Index (FMTI) in May 2024. The industry mean was 58. The index evaluates companies on how much they disclose about training data, model architecture, capabilities, limitations, and downstream use policies.
AI21 disclosed specifics including its 1.2 trillion token training corpus composition, data pipeline processes involving both human and machine evaluation, and distribution channels across AWS Bedrock, Google Cloud Marketplace, Snowflake, and Dataiku. That level of documentation exceeded what many larger competitors offered at the time. For those tracking AI privacy and transparency standards, AI21’s score is one of the highest among commercial LLM providers.
AI21 Jurassic Statistics: Enterprise LLM Market Share
The enterprise LLM market reached $6.7 billion in 2024 and was forecast to hit $8.8 billion in 2025, growing at a 26.1% CAGR through 2034. By mid-2025, enterprise LLM spending had already crossed $8.4 billion, according to Menlo Ventures.
AI21 holds roughly 0.01% of that market. The top seven companies — Microsoft, OpenAI, Anthropic, Google, AWS, Cohere, and AI21 — accounted for about 79% of the enterprise LLM market in 2024. Anthropic overtook OpenAI as the leading enterprise LLM provider by mid-2025, capturing 32% of production usage. OpenAI dropped to 25%, down from 50% at the end of 2023. Google claimed 20%. AI21’s position reflects the scale advantage held by companies with larger distribution networks and deeper AI workforce investments.
| Provider | Enterprise Market Share (Mid-2025) |
|---|---|
| Anthropic | 32% |
| OpenAI | 25% |
| 20% | |
| Meta (Llama) | 9% |
| Others (incl. AI21, Cohere, DeepSeek) | 14% |
Source: Menlo Ventures 2025 Mid-Year LLM Market Update
From Jurassic to Jamba: AI21’s Model Transition
AI21 released Jamba in March 2024 as its first production-grade model built on the hybrid Mamba-Transformer architecture. The Jurassic line, while still accessible through AI21 Studio, is now considered the company’s legacy product. As of August 2025, AI21’s product lineup consists of the outdated Jurassic family, the Jamba series, AI21 Studio, and the Maestro agentic planning system.
Jamba 1.5 Large and Mini launched in August 2024, with Jamba 1.6 following in March 2025 for private enterprise deployment. The Jamba models offer a 256K token context window — the longest among open-weight models — and process long contexts up to 2.5 times faster than competing architectures. Jamba Large and Mini outperform Mistral Large 2 and Llama 3.3 on general benchmarks, RAG benchmarks, and long-context question answering.
| Model Family | Architecture | Context Window | Status (2026) |
|---|---|---|---|
| Jurassic-1 (2021) | Transformer | Standard | Deprecated |
| Jurassic-2 (2023) | Transformer | Standard | Legacy / Available |
| Jamba (2024) | Mamba-Transformer Hybrid | 256K tokens | Active |
| Jamba 1.5/1.6 (2024–2025) | Mamba-Transformer MoE | 256K tokens | Active (Latest) |
Source: AI21 Labs, Contrary Research
AI21 Jurassic Pricing and Platform Availability
AI21 operates on a usage-based pricing model. For the current Jamba lineup, Jamba Mini costs $0.20 per million input tokens and $0.40 per million output tokens. Jamba Large runs at $2.00 per million input tokens and $8.00 per million output tokens. The Jurassic-2 models, still available via API, carry separate pricing tiers.
AI21’s models are distributed through multiple cloud platforms: Amazon Bedrock, Amazon SageMaker, Google Cloud Marketplace, Snowflake Snowpark, and Dataiku. Companies with over $50 million in revenue must purchase a commercial license for on-premises deployment. The breadth of these partnerships has not translated into proportional market share. Users exploring how AI tools are used across different platforms will find AI21’s deployment footprint is broad but thin compared to leaders.
Nvidia Acquisition Talks and Nebius Rumors
In late December 2025, Calcalist reported that Nvidia was in advanced talks to acquire AI21 Labs for $2–$3 billion. The deal was widely characterized as an “acquihire” — Nvidia’s primary interest was AI21’s workforce of roughly 200 employees with advanced degrees and rare LLM development expertise.
AI21 denied the acquisition talks in January 2026. Co-founder Amnon Shashua told Bloomberg that discussions had occurred with Nvidia and other companies, but nothing was close to completion. By April 2026, The Information reported that Nebius — an AI infrastructure company backed by Nvidia — had entered acquisition talks with AI21 after the Nvidia deal collapsed. Nebius had previously acquired Israeli startup Tavily for up to $400 million. Financial terms of the Nebius discussions have not been disclosed.
AI21 Jurassic Statistics: Competitive Landscape
AI21 competes with OpenAI, Anthropic, Google, Meta, Mistral, and Cohere in the enterprise LLM space. The company was absent from foundation model lists published by a16z, McKinsey, Lightspeed Ventures, and Menlo Ventures as of mid-2025. Cohere, a direct competitor in the enterprise-focused LLM category, reached $150 million in ARR by October 2025 — nearly three times AI21’s revenue.
AI21’s technical differentiator is its Mamba-Transformer hybrid architecture, which delivers 10x more memory efficiency than Llama-3 or Mistral and 30% lower compute requirements. The Jamba models also support structured JSON output, function calling, and citation generation natively — features that enterprise buyers increasingly expect. Tracking the broader AI governance and compliance landscape is relevant here, as AI21’s high transparency scores may appeal to regulated industries.
| Company | Est. Revenue (2024–2025) | Total Funding | Key Model |
|---|---|---|---|
| OpenAI | $13B ARR (Aug 2025) | $30B+ | GPT-4/5 |
| Anthropic | $5B ARR (Jul 2025) | $15B+ | Claude 4 |
| Cohere | $150M ARR (Oct 2025) | $1B+ | Command R+ |
| AI21 Labs | $57.8M (2024) | $636M | Jamba 1.6 / Jurassic-2 |
Source: Sacra, Menlo Ventures, Contrary Research
AI21 Jurassic Statistics: Key Company Milestones
AI21 was founded in November 2017 in Tel Aviv by Yoav Shoham, Ori Goshen, and Amnon Shashua. Shashua also founded Mobileye, the autonomous driving company acquired by Intel. The team launched Wordtune, an AI writing assistant, in October 2020 — Google named it one of its favorite extensions of 2021.
Jurassic-1 arrived in August 2021 with 178 billion parameters and a vocabulary exceeding 250,000 tokens. It was, at the time, one of the largest language models available for developer use. Jurassic-2 followed in March 2023 with improved instruction-following and multilingual support. In March 2025, the company launched Maestro, an AI planning and orchestration system designed to improve the accuracy of models like GPT-4o and Claude 3.5 Sonnet on complex tasks.
FAQ
What is AI21 Jurassic-2?
Jurassic-2 is AI21 Labs’ second-generation large language model, released in March 2023. It comes in three sizes (Light, Mid, Ultra) and supports zero-shot instruction-following and multilingual text generation.
How much funding has AI21 Labs raised?
AI21 Labs has raised approximately $636 million across seven rounds, including a $300 million Series D in May 2025 led by Nvidia and Alphabet.
Is Jurassic-2 still available in 2026?
Yes. Jurassic-2 remains accessible through AI21 Studio’s API, though the company now focuses development on its newer Jamba model family.
What replaced the Jurassic models at AI21?
AI21’s Jamba series, built on a hybrid Mamba-Transformer architecture, replaced Jurassic as the company’s primary model line starting in March 2024.
Is Nvidia acquiring AI21 Labs?
Nvidia’s reported $2–$3 billion acquisition talks fell through in early 2026. As of April 2026, AI infrastructure company Nebius is in separate acquisition discussions with AI21.
Sources:
https://research.contrary.com/company/ai21-labs
https://en.wikipedia.org/wiki/AI21_Labs
https://crfm.stanford.edu/fmti/May-2024/company-reports/AI21_Jurassic-2.html
