Loading…
Your privacy choices
Necessary storage keeps the site working. Analytics and advertising stay off unless you allow them. Cookie notice
AI21 Labs develops the Jamba family of foundation models, built on a hybrid Joint Attention and Mamba (Mamba-Transformer) architecture. The model lineup includes Jamba Large (398B parameters, 94B active), Jamba Mini (52B parameters, 12B active), and Jamba 3B, all featuring a 256K token context window. Jamba models are open-weight, privately deployable in a VPC or on-premises, officially support nine languages (English, Spanish, French, Portuguese, Italian, Dutch, German, Arabic, and Hebrew), and accept text input and produce text output. The company states its mission as supercharging human productivity with machines working alongside humans as thought partners. AI21 holds SOC 2 compliance and ISO 27001, 27017, and 27018 certifications, and offers the models via the AI21 Studio API.
This data may be out of date
Last checked 10 Nov 2026, 09:16 UTC. A newer update is expected. Records released or published after this time are not shown.
Every model version this company develops, newest first; each links to its own model card.
1–4 of 4
| Model | Released | Context window | Weights | Licence |
|---|---|---|---|---|
| Jamba Mini | 256,000 tokens Per models.dev; not confirmed by the developer | Open weights Per models.dev; not confirmed by the developer | ||
| Jamba Large | 256,000 tokens Per models.dev; not confirmed by the developer | Open weights Per models.dev; not confirmed by the developer | ||
| AI21-Jamba2-Mini | 262,144 tokens Per Hugging Face; not confirmed by the developer | Open weights Per Hugging Face; not confirmed by the developer | ||
| Jamba Mini 1.7 |
Offers this company serves as a provider, whoever developed the model. Each price uses the currency and billing unit shown with it.
1–3 of 3
| Model | Input, USD / 1M | Output, USD / 1M | Updated |
|---|---|---|---|
| Jamba Large | 2.00 | 8.00 | |
| Jamba Mini | 0.20 | 0.40 | |
| Jamba Mini 1.7 | 0.20 | 0.40 |