Jamba 1.5 Mini

**Jamba 1.5 Mini** is an efficient, powerful, instruction-tuned Foundation Model. It offers a 256K context window that’s optimized for long-form input, speed and cost effectiveness with 12B active parameters and 52B total parameters.

Jamba 1.5 models are released under the Jamba Open Model License, a permissive license allowing full research use and commercial use under the license terms. If you need to license the model for your needs, [talk to us](https://www.ai21.com/talk-to-us).

**Key features:**

- Optimized for Speed: Jamba's latency outperforms all leading models of comparable sizes.
- Long Context Window: Jamba's 256k context window is the longest available in any open model and [effectively delivers on the promise of long context](https://www.ai21.com/blog/long-context-yoav-shoham).
- Novel Hybrid Architecture: Jamba's Mamba-Transformer MoE architecture delivers cost and efficiency gains. With this architecture, developers can handle context lengths up to 140K tokens on a single GPU.
- Better for Developers: Jamba offers key features out of the box, including function calling, JSON mode output, document objects and citation mode.
- Multilingual: In addition to English, the model supports Spanish, French, Portuguese, Italian, Dutch, German, Arabic and Hebrew.

See [model card](https://huggingface.co/ai21labs/AI21-Jamba-1.5-Mini) for additional details and quality benchmarks.

**Use cases:**
Ideal for long-context use cases that require speed and cost efficiency. Use Jamba 1.5 Mini for fine-tuning and RAG applications. Sample use cases include:

- Digital assistants powered by extensive internal knowledge bases
- Research and analysis to gain insights from long, complex reports, such as multiple years of SEC filings
- Multi-doc comparison, for example reconciling healthcare claims with coverage policies
- Long, multi-turn chats in which the model needs to remember the full context of the conversation in order to generate accurate responses

We can't wait to see what you build with Jamba!