4 sources
A worthwhile head-page upgrade because it gives one of the quieter Jamba contributors a concrete place in the stack: the pre- and post-training work that turns a hybrid architecture into an actual usable model.
Researcher Profile
Editor reviewedMatan Danos
Hybrid Transformer–Mamba language models (Jamba)
Contributor to AI21's pre- and post-training work on Jamba
Useful because it captures one of the less-visible people behind AI21’s training stack, where hybrid-model quality depends as much on pre- and post-training choices as on the architectural headline.
Organizations
AI21 Labs
Labs
About This Page
This profile is meant to help you get oriented quickly: why this researcher matters, what to read first, and where to explore next.
Last reviewed
March 18, 2026
Known For
The ideas, systems, and research directions that make this person worth knowing.
01
Pre- and post-training work on Jamba
02
AI21 hybrid-model development
03
Turning model architecture into strong downstream behavior
04
Hybrid Transformer–Mamba language models (Jamba)
05
Jamba: A Hybrid Transformer-Mamba Language Model
06
Jamba-1.5: Hybrid Transformer-Mamba Models at Scale
Start Here
Canonical papers, project pages, or repositories that anchor this profile.
Supporting Sources
Additional links that help verify and flesh out this profile.
Related Researchers
People worth exploring next because they share topics, labs, or source material with this profile.
4 sources
A solid head-page upgrade because it turns another thin Jamba coauthor page into a real profile tied to pre- and post-training, the part of the stack where hybrid-model behavior gets tuned into something shippable.
4 sources
A useful page because evaluation work is easy to flatten into leaderboard noise, and her profile anchors the people inside AI21 who were responsible for turning Jamba performance claims into something measurable.
4 sources
A valuable systems page because hybrid-model launches depend on much more than modeling alone, and his contribution bucket points directly at the serving and infrastructure work needed to make Jamba usable in practice.
4 sources
A better long-tail AI21 page because it makes the data side of Jamba visible, instead of leaving the impression that hybrid-model progress came only from architecture and not from the people shaping the data pipeline underneath it.
4 sources
A good page to keep because it makes the project-and-product layer of a frontier-model launch visible; Jamba-1.5 was not only a research effort, it also needed people coordinating what got built, packaged, and released.