Zen 6 and Zen 6 Flash are the generation available now and Zen 7 is a research preview, so the header stops calling Zen5 "Latest" and the home stops presenting Zen 5 as the next generation. The home, models page and footer say who makes Zen LM, Zoo Labs Foundation, a 501(c)(3) non-profit, and the two jobs it is chosen for: agentic coding on your own machine, and marketing work. Every Zen 6 number is read from its model card. Upstream names appear only in the Architecture rows and the License & attribution section, both marked data-upstream. scripts/check-zen.mjs reads the built pages and fails the export when a name from zen.upstream shows outside those marks; pages.yml runs it over docs/ too. src/lib/blog.ts was never committed, because an unanchored "lib/" in .gitignore hid it, so a fresh clone could not build the blog. It is written again and the rule anchored. The datasets page drops the base-model column and the three rows that announced models still training or planned.
2.6 KiB
zen — the zenlm.org website
Repository: https://github.com/zenlm/zen · Site: https://zenlm.org
Zen LM is the open model family of Zoo Labs Foundation, a 501(c)(3) non-profit.
This repository is its website: a Next.js static export whose built pages are
committed under docs/.
The story the site tells
- Two jobs lead every overview: agentic coding that runs on your own machine, and marketing work.
- Zen 6 and Zen 6 Flash are available now (
zen6,zen6-flashon api.hanzo.ai; weightszenlm/zen6,zenlm/zen6-flash). Numbers come from those two model cards and nowhere else. - Zen 7 is a research preview: no weights, not callable. "Request access" goes to https://hanzo.ai/research-access.
- Zen 5, Zen 4 and Zen 3 are earlier generations. Zen 5.8 and zen6-coder are retired and are not listed.
Layout
src/app/— pages: home,/models,/datasets,/research,/blog.src/components/— Header, Footer, CatalogSection (renderssrc/data/catalog.json).src/lib/blog.ts— readscontent/blog/*.md(x)(gray-matter), lowers the four post components (Figure, LinkButton, Video, Fullwidth) to HTML and renders with unified/remark/rehype. It was missing from git until 2026-09-30 because an unanchoredlib/in.gitignorehid it; that rule is anchored now.content/blog/— a copy of github.com/zenlm/zen-blogcontent/. Keep the two in step: a post fixed there is copied here.docs/— the committed export..github/workflows/pages.ymluploads it as is; CI does not build.
Build
npm install
npm run export # next build → scripts/check-zen.mjs out → out/ becomes docs/
npm run typecheck
Zen copy names Zen
scripts/check-zen.mjs reads every built page the way a reader meets it (title,
meta and social tags, text, alt) and fails on a name from zen.upstream outside a
data-upstream element. Upstream names appear in two marked places only: the
Architecture value (the loader string) and the License & attribution section. A
post whose subject is another lab's release sets upstream: true in its
frontmatter; its page then carries <meta name="data-upstream"> and its article
and index card carry the attribute. It runs in npm run export and again in
pages.yml over docs/.
Serving
zenlm.org is served by the edge's staticFiles middleware from
s3://hanzo-sites/zen/zenlm (universe infra/aws/routes/sites.yaml, route
zenlm-org), a copy of the Pages build. GitHub Pages still deploys on push, but
no DNS name points at it: a change reaches zenlm.org only when docs/ is
published into that prefix.