Vercel AI Updates: October 27, 2026

1. GPT-6.1 Terra Arrives on AI Gateway

Vercel added OpenAI’s GPT-6.1 Terra to AI Gateway as openai/gpt-6.1-terra, with a fast mode at openai/gpt-6.1-terra-fast. Vercel describes it as a fast model for visual reasoning, computer use and writing, aimed at high-volume document and desktop-app work such as form filling, reading technical drawings and invoice entry. OpenAI still recommends its larger GPT-6 models for the hardest coding tasks. The model takes text and images, has a context window of more than 1M tokens, and supports configurable reasoning effort, structured outputs, tool calling and prompt caching. It works through the AI SDK, the Chat Completions and Responses APIs, and coding agents set up with npx vercel ai-gateway setup. Source

2. Vercel Agent Validates Alerts Before They Reach the Pager

Vercel now lets Vercel Agent investigate alerts from Vercel monitors before they go to a paging system. With “Critical Only” set on a webhook destination such as incident.io, only alerts the Agent classifies as Critical are forwarded. The Agent checks logs and metrics to judge impact, for example separating health-check errors that do not affect customers from sustained 5xx errors on checkout. Teams can add per-monitor context to guide it, and the CLI supports vercel monitors notifications --critical-only. The feature requires Pro or Enterprise with Observability Plus, which includes 10 Agent investigations per billing cycle, and custom monitors cost $0.0035 per GB scanned. Source

3. Scale AI’s AgentEnv Evals Run on Vercel Sandbox

Vercel announced that AgentEnv, Scale AI’s framework for testing agents against tasks and scoring the results, now runs on Vercel Sandbox. The Vercel provider ships in AgentEnv 0.9.1334 and later and runs the agent and its task environment in separate Sandboxes. An example repository runs an evaluation with the existing Vercel CLI login and shows the task score along with the agent’s recorded actions. Source

4. Vercel Connect Service Discovery Comes to the CLI

Vercel added vercel connect discover and vercel connect categories to the CLI so developers and agents can search more than 1,000 Vercel Connect services from the terminal. Search covers name, alias, keyword and description. Results can be filtered by category, by connection method such as mcp or oauth, or to services that forward webhook triggers. Both commands support --format=json for scripts and agents, and the feature requires CLI 64.1.0 or later. Source

5. Vercel Sandbox Commands Accept Standard Input

Vercel Sandbox now supports sending input to running commands over stdin, which makes it possible to talk to stdio MCP servers such as agent-browser inside a Sandbox. In the TypeScript SDK, runCommand() takes stdin: true and detached: true, with writeStdin() and closeStdin() to drive input, or a Node.js readable stream such as process.stdin. The Python SDK uses stdin=subprocess.PIPE with stdin.send() and communicate(). Source