Introducing Geist Serif
Geist Serif is a high-contrast display face for headlines, available now on npm and as a download in Regular weight, with more weights to follow.
Every Vercel update in one feed — release notes, features and changes, straight from the source.
1309 updates
Geist Serif is a high-contrast display face for headlines, available now on npm and as a download in Regular weight, with more weights to follow.
Daemons in Public Beta on all plans can now point a livenessProbe at an HTTP endpoint, and Vercel restarts the container when it fails.
InclusionAI's 1M-token coding and agent model costs nothing on AI Gateway until November 3, 2026, and works in Claude Code, Codex and Cursor.
Pro projects get 8x the cache storage and Enterprise 10x, but reads and writes that were unbilled under Data Cache will now be charged.
Pro projects get 8x the cache storage, but previously unbilled reads and writes now cost from $0.40 and $4.00 per million units.
Daemons declare a binding that names a service and an env variable, then call the private service's URL with a plain HTTP request.
Existing mistral/mistral-large-4 calls on AI Gateway pick up the new weights and lower token prices automatically, with no change to the model ID.
Enabled for all Vercel projects, the update gives postal codes for 90% of traffic, and VPN and hosting-provider visitors may now resolve elsewhere.
The Vercel Plugin adds Vercel API guidance and skills for AI SDK, Functions and Workflow, installed from Customizations → Marketplace in Antigravity.
The AI Bots managed ruleset lets you set separate log or block actions for crawlers and assistants, from the dashboard or CLI, on all plans.
Updating @flags-sdk/vercel to 1.5.3 cuts p75 cold-start initialization from over 70 ms to 10 ms and stops usage reporting from adding function time.
Connect adds webhook triggers for Resend, X, Twilio, Notion and Telegram, joining Slack, Discord, Teams, GitHub and Linear.
Strix users can set one AI Gateway key and pick from 400+ models, with usage and cost tracked in Vercel and no markup.
Projects whose engines.node range already allows 26, such as >=20, move to it on their next deployment, and HTTP/2 fetch can be switched off.
T3 Code users can add fx as a provider on web, desktop and mobile, with Supervised and Auto modes and their existing fx models and skills.
Box Mount is in private preview, with access requested from Box, and syncs a Box folder to a sandbox directory so agent output survives sandbox deletion.
Vercel CI runs only the Turborepo tasks a change affects, is open to all plans, and cut push-to-pass time 2-6x in Vercel's own repositories.
The Beta runs only Turborepo tasks tagged "ci", with Remote Cache on by default, and currently supports GitHub repositories on Linux.
Vercel CLI 64.2.0 drops the interactive-terminal requirement, though Observability Plus still needs an existing team subscription.
Layers in Vercel's registry can now reach 2 GB compressed, up from 500 MB, while the total image limit stays at 15 GB.
Particle-1, post-trained on DeepSeek V4.1 Flash with a 1M-token context, is callable as particle/particle-1 and works in Claude Code, Codex and Cursor.
The plugin needs Paseo 0.9 and Node.js 24, and stopped sandboxes keep files and history as snapshots that expire after 24 hours.
Teams can swap models, fallbacks and Zero Data Retention behind one ID in AI Gateway, with apps and coding agents picking changes up in minutes.
Developers can mix text and image file parts in one message, and all ten harness adapters, from Claude Code to Pi, process them with their vision models.
The 67% average saving applies to dashboard and Slack sessions with no action needed, and billing rates stay the same.
Vercel's AI Gateway now carries 14 ElevenLabs speech, music and sound effect models, at half price through November 9th.
The new Observability tab breaks down AI crawler and assistant requests, and CLI queries for the same data require Observability Plus.
Vercel's beta Container runtime builds a root Dockerfile or Containerfile on all plans, with Active CPU billing and no other framework detected.
The plugin needs OpenCode 2 and adds Vercel skills, specialist agents, and deploy and env commands, with the Vercel MCP server connected separately.
The AI SDK's experimental fx adapter keeps model credentials on the host and adds tool filtering, mid-turn steering, structured output and resumable sessions.