The problem
Enterprises want generative AI grounded in their own documents, but general chat models invent facts, cannot cite sources and charge per token at scales that break IT budgets — while training bespoke models in-house is out of reach for nearly everyone.
How it works
An occupier-company founded in Tel Aviv (Occupied Palestine) in November 2017 by Yoav Shoham, Ori Goshen and Amnon Shashua, selling to global enterprises — not a Palestinian company. AI21 Labs launched the Wordtune writing assistant in 2020, then Jurassic-2 task-specific models engineered for distinct language capabilities such as grounded question answering over proprietary datasets. Its Maestro system sits above the models: it plans multi-step work, routes each step to the right model — including open-source ones — and verifies outputs against source data, so companies get agents that cite rather than invent.
Pain points
Hallucinations in customer-facing answers, black-box models with no audit trail, token bills that explode at production scale, and pilots that stall on messy enterprise data.
Business model
Usage-based API pricing for models, per-seat subscriptions for Wordtune, and enterprise platform contracts for Maestro — landing with developers and expanding into department-wide deployments.
Challenges
Matching OpenAI, Anthropic and hyperscaler foundation models on capability-per-dollar, proving Maestro’s reliability gains on independent customer benchmarks, and covering training costs while the market consolidates around a few giant labs.
Funding
- Raised: $208 million Series C closed in November 2023 at a $1.4 billion valuation, adding Intel Capital and Comcast Ventures to the round.
- Valuation: $1.4 billion (November 2023).
Latest — November 2025
On 26 November 2025 AI21 Labs and Together AI announced a strategic partnership connecting Maestro, AI21’s agent orchestration system, with Together AI’s open-source model platform — letting companies build knowledge agents powered by optimised open models. “Maestro was built to orchestrate models intelligently, regardless of origin,” said co-CEO Ori Goshen, marking AI21’s shift from standalone models to model-agnostic enterprise orchestration.