Jamba: the most powerful & efficient long context models.

Open for builders, built for the enterprise.

Start Building on AI21 Studio

Try it on Hugging Face

Jamba 1.5 Mini & Large:
A Family of Open Models

Optimized for Speed

Jamba's latency outperforms all leading models of comparable sizes.

Longest
Context Window

Jamba's 256k context window is the longest openly available.

Novel Hybrid Architecture

Jamba's Mamba-Transformer MoE architecture is designed for cost & efficiency gains.

Better for Developers

Jamba offers key features OOTB including function calling, JSON mode output, document objects and citation mode.

Learn more about Jamba

2.5X faster than
all leading competitors

Jamba 1.5 models are optimized for speed.

Long context that actually delivers

Jamba 1.5 models maintain high performance across the full length of their context window.

Intelligent across the board

Jamba 1.5 models achieve top scores across common quality benchmarks.

Jamba 1.5 Availability

Choose the right model for the job

Choose the right state-of-the-art language model to power your GenAI workflows.

Our most powerful
FM to date

Jamba Large receive top scores across common benchmarks, using its hybrid SSM-Transformer architecture to blend accuracy with efficiency.

Tackle any business task

Specifically engineered to outperform on common business tasks, Jamba Large accompanies any team.

learn more

Explicitly enterprise

Unlike general purpose Foundation Models, Jamba Large is built especially for the enterprise, trained to excel on 25 core tasks that all enterprises rely on every day.

Precise FM functionality

With a 256K context window, Jamba-Instruct enables enterprises to more reliably and accurately summarize and query lengthy documents or databases.

Best cost per value

Specifically engineered to outperform on common business tasks, Jamba Large can...

learn more

Context length parity

Jamba-Instruct is one of the few language models on the market to maintain top performance across the length of its entire context window, with no drop between its claimed and effective context lengths.

Start Building