AI21 Labs builds foundation models and AI systems for the enterprise. Its Jamba model family, Jamba Mini and Jamba Large, is available through APIs and an SDK on usage-based pricing: Jamba Mini costs $0.2 per million input tokens and $0.4 per million output, while Jamba Large costs $2 and $8. Beyond models, AI21 offers an Intelligent Gateway, a drop-in endpoint that applies routing, caching and compaction to cut token waste based on actual agent traffic, plus dedicated small models tuned to match frontier quality on a customer's own workloads. New accounts get $10 in credits for seven days with no credit card.
The audience is enterprise AI and platform engineering teams, machine learning engineers and technical leaders who run agents or LLM workloads at a volume where token spend becomes a line item. Companies that need long-context models, private cloud hosting or a dedicated account manager can move to a custom plan with volume discounts and higher rate limits. Smaller developers can start on pay-as-you-go with unlimited seats and no commitment.
Teams use the Gateway by pointing an existing agent at it after a two-line setup, after which routing and caching start working on live traffic, and AI21 reports that most teams see spend drop within a day. An optimizer for agent configurations, listed as coming soon, will search configurations against private evaluations to meet a cost ceiling and quality floor. For stable, high-volume workloads, AI21 distills a dedicated model that keeps the accuracy of a larger frontier model at a fraction of the price. Research notes on the company's pages cover cost-quality tradeoffs between open and frontier models.
