Vercel AI Updates: November 17, 2026
Fully automated. Found, summarized and published by an automated AI pipeline. How posts are labeled
1. AI Gateway Adds Private Inference for Fine-Tuned, Self-Hosted and Prerelease Models
Vercel added Private Inference to AI Gateway, which connects a team’s own model endpoints to the same authentication, SDKs, APIs and observability as catalog models. Endpoints can be called through Chat Completions, Responses or Messages, with AI Gateway translating each request to the endpoint’s format. Private models are visible only to the owning team, credentials are stored encrypted, Zero Data Retention can be enabled per provider, and a catalog model can be set as a fallback. Providers can be registered in the dashboard or with the Vercel CLI 65.16.0 or later. Source
2. Tencent’s Hy4 Flash Arrives on AI Gateway at 50% Off Through November 29
Vercel made Hy4 Flash from Tencent available on AI Gateway as tencent/hy4-flash, with an introductory 50% discount on input, cached input and output tokens through November 29, 2026. Vercel describes it as the lower-cost model in the Hy4 family at roughly a tenth of the price per token, keeping reasoning with preserved thinking across turns, tool calling, structured outputs, prompt caching, image input and a 1M-token context window. Reasoning can be turned off for requests that do not need it. Source
3. Vercel Lays Out an Agentic Testing Pyramid for Evaluating Software That Agents Use
Vercel published a 14-minute guide arguing that agents are now users of APIs, CLIs, MCP servers and SDKs, so those products need evals. It groups tests by how many LLM calls they need: deterministic tests with none, one-turn evals with one, and end-to-end evals with two or more across a whole task. The post says the agent funnel has four stages (discovery, selection, access and success) and focuses on success, where an agent can complete a “successful” request that is still the wrong outcome, such as ordering a hot latte when an iced one was asked for. It recommends repeating tasks and grading outcomes before and after a change. Source