Hy4 is the first model in the Hy line to support images, and also brings computer use and a 1-million-token context window. See what changes in practice when testing the model.
An experiment published on CSS-Tricks uses two new CSS functions, if() and sibling-index(), to recreate the logic of :nth-child() with variables that change in real time, without touching the selector.
Vercel added an interactive card to its MCP server that shows a deployment's status, logs, and commit directly inside ChatGPT and other MCP Apps-compatible clients, without needing to open the dashboard.
Meituan's multimodal model has joined the Vercel AI Gateway catalog, with a 1 million token context window for code, documents, and long agent workflows.
The Vercel Plugin, available in the Antigravity marketplace (Google's agentic IDE), gives AI agents guidance on Vercel APIs and patterns, including an example Python backend deployed via Vercel Sandbox.
The company is bringing Data Cache inside Runtime Cache, unifying four caching APIs under the same infrastructure. Developers using fetch or unstable_cache in Next.js don't need to touch their code, but the storage and billing model changes underneath.
The AI SDK's HarnessAgent layer now processes images alongside text across any of the ten supported adapters, from Codex to GitHub Copilot. The feature removes the need for workarounds to give visual context to agents running in a sandbox.
Vercel announced in the platform's changelog that Vercel Agent sessions in the dashboard and on Slack became, on average, 67% cheaper, with no change to billing rates. The gain came from prompt caching and context management optimizations that any team building agents can replicate.
Strix, an open source pentesting tool for AI systems, now runs on top of the Vercel AI Gateway, trading multiple provider credentials for a single key and one cost dashboard.
New AI Gateway layer separates the model ID used in code from the actual configuration behind it, allowing you to swap model, fallback, and retention policies without a deploy.
A plugin connects Paseo to Vercel Sandbox: the conversation with the agent is saved in the cloud, you can switch devices mid-task, and the AI Gateway key never enters the execution environment.
When no framework is detected, Vercel now builds the Dockerfile or Containerfile at the project root and routes all traffic to the resulting image. See the priority order between framework, container file, and the manual flag.
Vercel now reuses pnpm's supply-chain verification across deploys. In cases with restored cache, installs that used to take seconds now finish in milliseconds.
Vercel's MCP server has added submit_feedback, which lets agents and copilots send bugs, friction points, and feature requests with structured context, without going through a form or a GitHub issue.
Vercel added Kimi K3.1 Pro and the Highspeed variant to AI Gateway, a model from China's Moonshot AI focused on coding, with 1 million tokens of context and weights expected to open by November 4.
