Vercel AI Updates: October 29, 2026

1. AI Gateway Adds Virtual Models to Switch Models Without Redeploying

Vercel introduced Virtual Models on AI Gateway, saved configurations called through an ID such as vmc/coding-agent in place of a provider model string. The model, provider order, fallbacks, reasoning effort, fast mode, prompt caching, Zero Data Retention and reporting tags are set centrally, and every app or agent using the ID picks up changes within a few minutes with no code change or redeploy. Settings on the virtual model override the same settings in a request, so callers cannot change the fallbacks or service tier, and ZDR or no-training requirements are enforced even if a request sets them to false. Provider-specific options stop applying when the virtual model is pointed at a different provider. Vercel suggests pointing a second ID, such as a -candidate variant, at a new model and running evals before switching production traffic. Coding agents such as Claude Code and Codex can select virtual models after vercel ai-gateway setup. Virtual Models work with every model type on AI Gateway and with the AI SDK and the OpenAI- and Anthropic-compatible APIs. Source

2. Vercel CI Enters Beta to Keep Up With Agent-Generated Pull Requests

Vercel launched Vercel CI in beta on all plans, pitched at the growing volume of pull requests from coding agents. It reads turbo.json, so Turborepo users tag tasks with "tags": ["ci"] instead of writing workflow YAML. On each push it runs only the tasks affected by the change, serves unchanged tasks from Remote Cache, and runs each remaining task on its own machine on Vercel’s Hive compute layer as soon as its dependencies finish. Vercel says time from push to “CI passed” in its own repositories is 2-6x faster, and the system has completed nearly 8 million tasks. Agents can inspect runs through the dashboard, CLI or Vercel MCP, and results post as GitHub checks for branch protection. Planned additions include a vc ci command for distributed execution before pushing and more granular test selection. Source

3. Particle-1 Coding Model Arrives on AI Gateway

Vercel added Particle AI’s Particle-1 to AI Gateway as particle/particle-1. The model is post-trained on DeepSeek V4.1 Flash to complete software engineering tasks with fewer tokens, which Particle attributes to shorter reasoning, fewer trial-and-repair rounds and more focused exploration. It has a 1M-token context window and can be selected in Claude Code, Codex, Cursor and other agents after running the Vercel CLI setup. Source

4. Box Mount Gives Sandbox Agents Filesystem Access to Box Folders

Vercel said Box Mount now runs in Vercel Sandbox, syncing a Box folder with a directory in the sandbox so agents can read and write Box files with ordinary filesystem operations. Files the agent writes remain in Box after the sandbox is deleted, as long as the folder is unmounted first. The sync process and the agent run as separate sandbox users, keeping Box credentials out of the agent’s workspace. A contract-review example uses AI Gateway to review a synthetic contract and save the result back to Box. Box Mount is in private preview. Source

5. Paseo Runs fx, Codex, Claude Code, OpenCode and Pi in Vercel Sandbox

Vercel released a Vercel Sandbox plugin for Paseo that runs fx, Codex, Claude Code, OpenCode and Pi in the cloud. A session started from Paseo on a laptop can be controlled from a phone, even with the laptop closed, and stopping the sandbox saves files and conversation history (active processes stop). The AI Gateway key is entered once and kept outside the sandbox. Source

6. fx Coding Agent Harness Becomes a Provider in T3 Code

Vercel made fx, its open-source coding agent harness, available in T3 Code, an open-source control plane for coding agents on web, desktop and mobile. The built-in fx provider supports Supervised and Auto modes along with the user’s existing fx models and skills, and is added under Settings > Providers. Source