What Vercel Shipped in Mid-September.

Eight changes landed on Vercel this week, split between new voice models on AI Gateway and speed improvements to builds and deployments. Builders running voice agents, custom environments, or static sites all get something useful here.

Gemini 3.8 Live

Vercel's AI Gateway now supports Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking from Google. The Live model handles real-time audio, visual grounding, automatic switching across 97 languages and background tool calls, while Extended Thinking adds multi-step reasoning that runs in parallel with speech, both reachable via the AI SDK's realtime API over WebSockets.

Is Agentic Site-Type Views

Is Agentic now lets you view audit results through one of four lenses: Docs & content, Business, App or Commerce. Each view surfaces checks relevant to that site type, the underlying score stays the same across views, and a new meta tag lets site owners set a default view without rescanning.

Faster Secure Compute And Static IP Builds

Builds using Secure Compute and Static IPs now start 64% faster, with average time from deployment creation to build start dropping from 6.7 seconds to 2.4 seconds thanks to prewarmed build containers. No configuration changes are needed.

GPT-Live 1

GPT-Live 1 from OpenAI is now on AI Gateway, a full-duplex voice model that listens and speaks at the same time without turn detection, so users can interrupt naturally. It also supports client delegation, letting developers hand deeper processing to a separate text model on AI Gateway while keeping the voice conversation running, and requires AI SDK 7.

Custom Environments For Marketplace Integrations

Native Marketplace resources can now connect to custom environments, not just production, preview and development. Set-up works through the dashboard, CLI or REST API, scopes environment variables accordingly, and is available on Pro and Enterprise plans; existing deployments need a fresh deploy after the environment scope changes.

Turbo Build Machines Per Deployment

Developers can now enable Turbo build machines on any single deployment without touching project settings, using a Git commit flag, vc deploy --turbo on CLI 59.20.0+, or the buildMachine field in the REST API, for temporary per-deployment speed.

Sub-Second Artifact Deployments

Vercel CLI 59.16.0+ now supports sub-second artifact deployments, skipping the build step entirely for eligible static sites and returning a live URL in under a second. It covers directories with up to 10 HTML or Markdown files totalling 5MB or less.

GLM 5.3 FlashX

Z.ai's GLM 5.3 FlashX is now on AI Gateway, a high-speed multimodal coding model running at roughly 200 tokens per second, aimed at coding agents and tool loops. It is accessible via the API format zai/glm-5.3-flashx with no markup on provider pricing.