Jamba Models by AI21 Labs
About
The Jamba family of large language models, developed by AI21 Labs, features two primary models—Jamba 1.5 Mini and Jamba 1.5 Large. These models leverage a novel architecture that integrates Transformer and Mamba layers to optimize speed, memory efficiency, and overall quality 128. One standout feature is their extensive context window of 256K tokens, allowing for superior handling of tasks requiring long text inputs, such as document summarization and analytical tasks 34. The Jamba models exhibit unparalleled performance in their class, outperforming competitors across various benchmarks 23. These models also offer multilingual support, accommodating languages such as English, Spanish, French, Portuguese, and more 11. They are readily accessible via AI21 Studio and multiple cloud platforms 9.
Current Variants
Use-when guidance is derived from seed capabilities, context, release, and replacement fields.
| Model | Use when | Released | Signals | Status |
|---|---|---|---|---|
| Jamba-Instruct | Use when the workload needs 256k context. | 2024-05 | 256k context | Current |
| Jamba v0.1 | Use when the workload needs 256k context. | 2024-03 | 256k context | Current |
Release Timeline
2 release groupsSpecifications(2 models)
| Model | Released | Context | Parameters |
|---|---|---|---|
| Jamba-Instruct | 2024-05 | 256k | 52B (12B active) |
| Jamba v0.1 | 2024-03 | 256k | 52B (12B active) |
Available From(2 providers)
Pricing
| Model | Provider | Input / 1M | Output / 1M | Type |
|---|---|---|---|---|
| Jamba-Instruct | AI21 Studio | $0.5 | $0.7 | Serverless |
| Jamba-Instruct | Microsoft Foundry | $0.5 | $0.7 | Serverless |
Frequently Asked Questions
- What is Jamba used for?
- Jamba is used for structured outputs, chatbot and role-playing use cases, and long-context generation. The family description and listed model capabilities point to those workloads as the best fit.
- How does Jamba compare to Jamba 1.7?
- Jamba by AI21 Labs is strongest where you need structured outputs, while Jamba 1.7 by AI21 Labs is the closest related family to check for agent workflows and tool use. Jamba has 2 listed variants and reaches up to 256k context, while Jamba 1.7 reaches up to 256k context, so compare the specs and pricing tables before choosing a production model.
- Which Jamba model should I use?
- For the lowest listed input price, start with Jamba-Instruct through AI21 Studio at $0.5/1M input tokens. For the most capable/latest local choice, evaluate Jamba-Instruct with 256k context.
