As AI consultants based in Rosenheim, Germany, we recommend AI21 Jamba for enterprises that need to process extremely long documents. With a 256k token context window, you can analyze entire books at once.
Important note (as of June 2026): In May 2026, AI21 Labs executed a strategic pivot and laid off more than 60% of its workforce. The company is now focusing on its Maestro orchestration platform for enterprise AI agents and is scaling back the sale of Jamba models as a standalone API product. However, the open weights remain available under Apache 2.0 on HuggingFace and can be self-hosted. For new production projects relying on a hosted API, we currently recommend evaluating alternative providers or self-hosting the Jamba2 models directly.
Innovative Architecture
AI21 Labs is an Israeli AI company that developed a unique hybrid architecture with Jamba: the combination of Mamba (State Space Model) and Transformer.
Why Jamba for Enterprises?
- 256k Context: One of the longest contexts on the market
- Efficient: MoE architecture reduces resource requirements
- Open Models: Jamba 1.5 under the Jamba Open Model License, Jamba 1.6 and Jamba2 under Apache 2.0
- Multilingual: Strong support for many languages
- EU-Available: Via AWS Bedrock Frankfurt
Key Strengths
Mamba-Transformer Hybrid
Jamba combines two architectures:
- Mamba Layers: Efficient processing of long sequences
- Transformer Layers: Precise attention for details
- MoE: Mixture-of-Experts for efficiency
This combination enables:
- 3x faster inference on long contexts
- 2x less memory requirements
- Linear instead of quadratic scaling
Extremely Long Context
256,000 tokens mean:
- ~640 pages of text at once
- Entire books to analyze
- Complete codebases to understand
- Comprehensive legal contracts to review
Open-Source Option
Jamba 1.5 is available as open source:
- Jamba Open Model License
- Self-hosting possible
- Full control over data
- Community-driven
Hardware Requirements (Self-Hosted)
| Model | VRAM | Recommended GPU |
|---|---|---|
| Jamba2 Mini | 48 GB | A100 80GB |
| Jamba 1.5 Large | 160 GB | Multi-A100 |
| Jamba 1.5 Mini | 24 GB | RTX 4090 |
Comparison to Other Models
| Feature | Jamba2 | GPT-4 | Claude 3 |
|---|---|---|---|
| Context | 256k | 128k | 200k |
| Architecture | Hybrid | Transformer | Transformer |
| Open Source | Yes (1.5/1.6/2) | No | No |
| MoE | Yes | No | No |
Integration with CompanyGPT
AI21 Jamba can be integrated in CompanyGPT - ideal for enterprises with extensive document collections.
Our Recommendation
AI21 Jamba2 Mini as an open-weights model under Apache 2.0 remains our recommendation for self-hosted document analysis with very long texts. If you regularly work with very long documents (legal contracts, book manuscripts, code reviews) and need full data sovereignty, Jamba2 on your own infrastructure is an excellent choice.
Due to AI21’s strategic shift to the Maestro platform, we advise caution for API-centric new projects: either commit to self-hosting (HuggingFace, AWS Bedrock) or evaluate alternatives. For general cloud-API applications without a specific focus on long contexts, we recommend Google Gemini (up to 1M context) or OpenAI GPT.