Vercel CI is in beta: it runs only the checks your change touches
SummaryVercel's new CI reads your Turborepo task graph, skips tasks your push didn't affect or that are already cached, and runs the rest each on its own machine. Vercel says its own push-to-green time got 2-6x faster. It's GitHub-only and Linux-only for now.
· MODEL Editors · 2 min read
Some links may be affiliate links. Disclosure

Vercel announced Vercel CI in beta on October 28. The pitch is aimed at teams whose coding agents open a lot of pull requests: most CI systems rerun the whole workflow on every push, even for a one-line change, so the more agents push, the longer everyone waits. Vercel CI tries to run only what a change actually affects.
How it works. Vercel CI uses Turborepo, Vercel's build system for JavaScript and TypeScript repos, to read your task graph from turbo.json. You add "tags": ["ci"] to each task you want it to run, such as lint or test. There's no separate workflow YAML. On each git push, it works out which tagged tasks the change affects, restores anything already in Vercel Remote Cache (on by default), and runs the rest in dependency order, each in its own compute instance. It posts a check to your GitHub pull request, so existing branch protection rules can block merges when it fails. If you don't use Turborepo, Vercel says you can add turbo.json and the turbo package, including in single-package repos.
For agents. Runs and logs are available in the dashboard, the Vercel CLI (vc ci ls, vc ci inspect, vc ci logs, vc ci grep), Vercel MCP and Vercel's OpenAPI spec. Vercel's suggested setup is to ask your coding agent to run npx skills use vercel/ci@setup, which picks initial tasks and opens a pull request.
The numbers are Vercel's. Vercel says time from push to "CI passed" in its internal repositories is 2-6x faster than before, and that Vercel CI has run nearly 8 million tasks across its own repos and the private beta. MODEL has not tested it.
Pricing. Vercel CI bills compute in CPU minutes: $0.0035 per CPU minute on Pro, custom on Enterprise. Each task uses 2 CPUs by default and run bootstrapping uses 4. Vercel's example: 8 CPUs for 2 minutes 30 seconds is 20 CPU minutes, which works out to $0.07 at the Pro rate. Skipped and cached tasks don't use CPU minutes. The blog says the beta is available on all plans, but the pricing page lists only Pro and Enterprise rates.
Who it's for: teams already on Turborepo and GitHub, especially ones where agents push often and full CI runs are the bottleneck.
The caveats, from Vercel's docs. It's a beta and "features and configuration may change." GitHub is the only supported forge, git push is the only trigger, and Linux is the only base image. The repo must be connected to at least one Vercel project and not to projects on more than one Vercel team. It needs Turborepo 2.11.7 or later and Vercel CLI 64.3.1 or later. How much it skips depends on how your tasks are split in turbo.json.
Not sponsored. No affiliate links.
Sources
- Introducing Vercel CI (blog, Oct 28, 2026)
Vercel
- Vercel CI is now in beta (changelog)
Vercel
- Vercel CI (docs)
Vercel
- Vercel CI Pricing (docs)
Vercel
- Announcement post
@vercel on X
FREE · EVERY WEEK
The top AI tools this week, tested.
Every week MODEL's editors pick the AI tools worth your time, try them hands-on, and tell you what's good, what's not, and who it's for. Free, in your inbox.
Free. Unsubscribe anytime. Our picks are editorial and can't be bought.
Continue

Tool Update
OpenAI's Tibo: dots are getting GPT-6 Astra, GPT-6.1 Terra, and more
OpenAI's Tibo Sottiaux said the team is expanding the frontier models that power ChatGPT dots beyond today's Astra-only setup, adding GPT-6.1 Terra for speed and personality so more people can get a dot. Your dot will pick the model. It hasn't shipped yet.
· MODEL Editors

Hot Take
Hot take: sloppy AI code is a checking problem, not a model problem
Anthropic says Claude wrote the Rust port of its Messages API, starting with no Rust developers on the team, and that the safety net was shadow traffic, not line-by-line review. The same week, Veracode found about 4 in 10 AI coding tasks still produce code with a known vulnerability. Here is what was actually said, what the viral version gets wrong, and our take.
· MODEL Editors

Tool Update
Notion's week: its agent now makes Excel, Word, PowerPoint and PDF files, and edits say which agent did them
Notion's Halloween recap bundles four updates from Oct 27 to 30: an agent that builds real office files, speaker labels for Google Meet notes, bigger automatic offline downloads, and a 'Last edited by' label that names the agent or MCP client behind a change. Here is what each one does, and what Notion hasn't said.
· MODEL Editors