Zamba
Zamba is a model family from Zyphra18, with 2 assessed releases in this catalog. Licenses and availability are recorded for each release, not for the family; its releases use the Apache License 2.0.
Zamba is Zyphra's series of hybrid models that combine Mamba state-space layers with shared transformer attention blocks. Releases include Zamba-7B (April 2024); Zamba2 at 1.2B, 2.7B, and 7B parameters (July to October 2024), with instruction-tuned versions; and Zamba2-VL vision-language models at the same three sizes (June 2026).1234
- Website: Zyphra research index (external site: zyphra.com)
- Model hub: Zyphra on Hugging Face (external site: huggingface.co)
- Repository: Zamba2 standalone PyTorch implementation (external site: github.com)
- Documentation: Zamba2 in Hugging Face Transformers (external site: huggingface.co)
- Paper: Zamba2-VL technical report (arXiv 2606.00390) (external site: arxiv.org)
On AI on Hugging Face: Zyphra/Zyda
Releases assessed
| Release | Weights | Tier (USASI rubric v0.2) | License | Released |
|---|---|---|---|---|
| Zamba2-7B | Weights: Public | Model-disclosure tier (USASI rubric v0.2): Open-weight | Apache-2.0, Apache-2.0 | Oct 14, 2024 |
| Zamba2-VL-7B | Weights: Public | Model-disclosure tier (USASI rubric v0.2): Open-weight | Apache-2.0, Apache-2.0 | Jun 2, 2026 |
What it is useful for
Organization context
Provenance and derivatives
The Zamba language models were pretrained by Zyphra on open web text and code, including its own Zyda dataset; the Zamba2 models use the Mistral v0.1 tokenizer. Zamba2-VL builds on the Zamba2 language models and uses the vision encoder from Alibaba Cloud's Qwen2.5-VL.869711
- Derived from: Zyda (external site: huggingface.co) — Zyphra's open pretraining dataset (ODC-BY), assembled from seven open datasets.
- Derived from: Qwen2.5-VL vision encoder (external site: huggingface.co) — Vision backbone for Zamba2-VL only; Qwen is Alibaba Cloud's model family.
U.S. eligibility
Eligible · basis: U.S. headquarters
The Zamba models are developed and published by Zyphra, whose terms of use name Zyphra Technologies, Inc. at a San Francisco address; the Zamba2-VL report lists the authors' affiliation as Zyphra, San Francisco, CA. Eligibility of each release is assessed on the release record.105
Sources
This listing is not an endorsement, a safety assessment, or a federal approval.
Share or cite this page
- X (opens X)
- Bluesky (opens Bluesky)
- LinkedIn (opens LinkedIn)
- Reddit (opens Reddit)
- Hacker News (opens Hacker News)
- Email (opens your email app)
- RSS updates
Cite this page
APA style
USASI. (2026, September 29). Zamba. United States of America Superintelligence. https://unitedstatesofamericasuperintelligence.com/open/zamba/
BibTeX
@misc{usasi_zamba,
author = {{USASI}},
title = {Zamba},
year = {2026},
month = sep,
howpublished = {\url{https://unitedstatesofamericasuperintelligence.com/open/zamba/}},
note = {United States of America Superintelligence. Last updated 2026-09-29}
}The date is when this page was last updated. Add the date you read it if your style needs one.
For maintainers: a “Listed in USASI” badge
Links to this page. It says the entry is listed, not endorsed or rated. About the badge
Markdown (for a README)
[](https://unitedstatesofamericasuperintelligence.com/open/zamba/)
HTML
<a href="https://unitedstatesofamericasuperintelligence.com/open/zamba/"><img src="https://unitedstatesofamericasuperintelligence.com/badges/listed.svg" alt="Listed in the USASI catalog" height="20"></a>