AI21 Jamba
AI Model
Language
Current
The production hybrid: Mamba state-space layers + transformer attention.
You're looking at an AI model record: who built it, what it can do, and the hardware and data behind it.
Jamba (2024–25) interleaves Mamba SSM blocks with attention to cut memory and speed long contexts — the most prominent deployed challenge to pure transformer orthodoxy, shipped as open models up to ~400B total parameters.
How this connects to the rest of the stack
This record's links at a glance — click a group to jump to it.
Developed by
Based on
Sources
Record reviewed as of 2026-01-15 — see Methodology.