Vercel — Partner Intelligence

Current state (as of 2026-05-17)

Next.js 16 with Turbopack is the default. AI SDK v6 is the current production line. AI Gateway went GA in August 2025 and is now the recommended model-routing layer in front of any production agent. Fluid Compute is the default compute mode for new projects — pay-per-active-CPU billing with sub-second scale-to-zero. The new vercel.ts (TypeScript config, replacing vercel.json) is shipping. Vercel Sandbox (isolated sandboxes for agentic code execution) went GA in January 2026. Vercel Queues is in beta. Vercel Agent (managed agent runtime) is in beta. Node 24 LTS is the default runtime. Default function timeout sits at 300 seconds on Fluid Compute.

Product map

• Next.js 16 — App Router, Turbopack default, Cache Components, Partial Prerendering (PPR), use-cache directive, async params in dynamic routes.
• Vercel AI SDK v6 — TypeScript-first model SDK. Unified surface over Anthropic, OpenAI, Google, xAI, Cohere, Mistral. Streaming, tool use, generateObject, agents.
• AI Gateway (GA Aug 2025) — model-routing layer. One API key, all providers. Caching, observability, fallback routing.
• Fluid Compute — default compute mode. Pay-per-active-CPU, sub-second scale-to-zero, 300s default timeout, in-function concurrency.
• Vercel Sandbox (GA Jan 2026) — isolated execution sandboxes for agentic code-execution workloads.
• Vercel Queues (beta) — managed durable queues for background work.
• Vercel Agent (beta) — managed agent runtime built on Sandbox + Queues.
• Vercel CLI — vercel, vercel deploy, vercel env, vercel link, vercel logs.
• Vercel MCP server — first-party MCP surface for deployment, logs, project info, runtime traces.
• vercel.ts config — TypeScript-typed deployment config, replacing vercel.json.
• Partner Program — Vercel Partner Program (agency + technology tiers). Entry criteria public on vercel.com/partners.
• Open-source artifacts — next (MIT), ai SDK (Apache 2.0), turbo/Turbopack (MPL-2.0), MCP server (MIT).