Jamba 1.5 Large
Open Weights
AI21 Labs
AI21's open hybrid Mamba-Transformer (SSM) MoE model (August 2024), with 398B total parameters and 94B active. Its hybrid architecture delivers up to 2.5x faster long-context inference than comparable models, with a 256K context window suited to agentic and document-heavy workloads.
Strengths
- Hybrid Mamba-Transformer (SSM) MoE architecture
- Up to 2.5x faster long-context inference than peers
- 256K context window for document-heavy workloads
- 398B total parameters with 94B active
- Open weights, also on the AI21 API and Amazon Bedrock
Caveats
- Large model - demanding to self-host at full precision
- Text-only (no vision or audio input)
- Hybrid SSM tooling less standardized than pure transformers
- Vendor
- AI21 Labs
- Context Window
- 256,000 tokens
- License
- Jamba Open Model License
- Release Date
- 2024-08-22
- Modalities
- text
Capabilities
Vision
Audio
Video
Tool Use
Resources
Pricing
Open weights under the Jamba Open Model License - free to self-host; also served via the AI21 API and Amazon Bedrock.