Vercel AI Updates: November 6, 2026
Fully automated. Found, summarized and published by an automated AI pipeline. How posts are labeled
1. Vercel Schedules Enters Public Beta as the Successor to Cron Jobs, With Schedules Agents Can Create at Runtime
Vercel launched Schedules in public beta on all plans, announced in a November 4 blog post and a November 5 changelog entry. It succeeds Vercel Cron Jobs (existing cron jobs keep working, and a migration guide covers the move) and runs tasks once or on a recurring cadence in production, preview or custom environments. Static schedules live in vercel.json or Build Output API configuration, sync on deploy, and can target a private Vercel Function, an HTTP path or a Vercel Queue topic. Adding a schedule/v1beta trigger makes a function private, with no public URL, so only Vercel’s scheduling infrastructure can invoke it. Dynamic schedules are created and managed at runtime through the JS SDK, CLI or API and deliver to a Queue topic, so an application or agent can set up one-time or recurring work without redeploying. Any schedule can be enabled, disabled or run on demand. Vercel’s example is an AI SDK tool that lets an agent save a request such as “Tomorrow morning, summarize the top five stories on Hacker News” as a schedule. When it is due, Queues deliver the saved request to the app, which checks the user’s current permissions and starts the agent; previously this took a database table plus a polling cron job. Schedules also powers dynamic scheduling tools in Vercel’s open-source eve agent framework and experimental scheduled tasks in Nitro projects. Pricing on Pro and Enterprise starts at $5 per million runs and $10 per million schedule creations, and Hobby includes 500 creations and 5,000 runs per month. Blog Changelog
2. Google’s Gemini Omni 1.2 Flash Video Model Arrives on AI Gateway
Vercel added Google’s Gemini Omni 1.2 Flash to AI Gateway as google/gemini-omni-1.2-flash on November 5, so it can be called with an AI Gateway key and no separate Google account. The model generates 3 to 15 second videos with native audio from text prompts and edits 3 to 15 second source clips with up to six reference images. It can extend a video by 3 to 10 seconds per call, up to 30 seconds in total, and refine generated clips in follow-up calls using previousInteractionId. Output comes in 360p, 720p, 1080p or 4K at 16:9 or 9:16. In the AI SDK, generateText or streamText returns the video as a file; with streamText, the clip arrives through fullStream as a completed file. AI Gateway passes through provider pricing with no markup and no platform fee on inference, including on BYOK requests. Source
3. Token Pass Credit Now Applies to Any AI Gateway Key and Falls Back to Team Credits When It Runs Out
Vercel changed Token Pass, the AI Gateway subscription for Pro teams it introduced in October at $30 a month with $80 of monthly credit for featured models, so that the credit applies automatically to included model usage whether a request authenticates with a standard API key, an OIDC token or fx login. A dedicated Token Pass key is no longer required, though one can still be created from the dashboard or CLI for keys scoped only to featured models and providers. When the credit runs out for the billing period, requests made with a standard key or OIDC token now fall back to the team’s AI Gateway credits, while dedicated Token Pass keys still stop until the next period. Teams that created a Token Pass key during setup therefore need to switch to a standard key or OIDC token to get the fallback. Requests served through the Token Pass provider remain Zero Data Retention, and fallback requests follow the team’s own setting, routing only to ZDR providers when team-wide ZDR is enabled. Source
4. Vercel Labs Ships AI SDK for Python 0.9 With a DurableAgent, and json-render 0.22 Renders Generative UI as Chat SDK Cards
Vercel Labs released v0.9.0 of the AI SDK for Python, which is in public beta, on November 4. It adds ai.durable.DurableAgent for agent loops that run inside durable workflow systems, where model and tool calls run as recorded steps that can be replayed after a restart; the repository includes examples for Temporal and Vercel Workflows. ToolRunner now tracks tool call IDs, deduplicates in schedule() and supports discard_all(), and new PendingHookEvent, ResolvedHookEvent and CancelledHookEvent types cover approval-style hooks. A native TypeSafe adapter can be created with ai.Model(id="jev-latest", provider=TypeSafeProvider()) until TypeSafe’s models are listed on models.dev. The same day, json-render v0.22.0, Vercel Labs’ Apache-2.0 generative UI framework (about 18,900 GitHub stars), added @json-render/chat, which renders specs as Chat SDK cards so one component catalog posts natively to Slack, Teams and other Chat SDK platforms. streamCard streams a card into a thread and edits it as elements arrive, toCard validates specs against per-component schemas and optional platform limits, and a playground previews messages as Slack renders them, with a Teams Adaptive Card view. AI SDK for Python json-render