Vercel AI Updates: October 28, 2026
Fully automated. Found, summarized and published by an automated AI pipeline. How posts are labeled
1. ElevenLabs Brings 14 Audio Models to AI Gateway
Vercel partnered with ElevenLabs to add 14 ElevenLabs models to AI Gateway, with all of them 50% off for the first 14 days, through November 9. The models cover six categories: speech to text, streaming text to speech, realtime voice agents, music, sound effects and multi-voice dialogue. Scribe v2 Realtime transcribes speech as the user talks, and Eleven Flash v2.5 can read a language model’s answer aloud while it is still streaming. Eleven Realtime v1 Max and Mini run a full spoken conversation over one WebSocket session, and the useRealtime hook in @ai-sdk/react handles the connection, microphone capture and playback using semantic voice activity detection. Music v2.5, Music v2 and Sound Effects v2 generate audio from text through generateSpeech, and Eleven v4, v4 Turbo and v3 voice multi-speaker scripts in a single request. Usage appears in the same cost and trace views as other AI Gateway traffic. Source
2. Vercel Agent Sessions Cost 67% Less on Average
Vercel said Vercel Agent sessions in the dashboard and Slack now cost 67% less on average after it rebuilt how the Agent manages context. The changes include workflow-based Vercel CLI skills with guidance on common workflows and error recovery (which other agents using the skills also get), incremental context management for long conversations such as Slack threads, better prompt-cache reuse across turns, and updated planning and coding guidance that uses fewer tokens. Billing rates are unchanged: paid sessions are billed at the model provider’s inference rates plus a $0.25 per million tokens Vercel Agent fee. Source
3. AI SDK fx Adapter Moves the Harness Out of the Sandbox
Vercel rewrote the fx adapter for the AI SDK’s HarnessAgent so the harness and model requests run in the same process as the application, while built-in filesystem and shell operations stay isolated in the sandbox. The adapter now uses libfx, the TypeScript library for fx, and no longer installs fx inside the sandbox, runs a sandbox bridge or exposes a sandbox port. Model credentials stay on the host instead of being forwarded into the sandbox. The rewrite adds built-in tool filtering, mid-turn steering, structured output, image tool results through toModelOutput(), and sessions that can be saved and resumed in a new host process. It ships in the latest @ai-sdk/harness-fx. Source
4. HarnessAgent Accepts Image Inputs Across All Adapters
Vercel added image input support to the AI SDK HarnessAgent layer for every harness adapter. Images are passed in the standard AI SDK message format as a content array mixing text and file parts, and are processed by each harness’s underlying vision model. Supported adapters are Claude Code, Cline, Codex, Cursor, Deep Agents, fx, GitHub Copilot, Grok Build, OpenCode and Pi. Source
5. Vercel Plugin Comes to OpenCode
Vercel made the Vercel Plugin available in OpenCode. The plugin provides on-demand guidance on Vercel APIs and recommended patterns, with skills for the AI SDK, Vercel Functions and Workflow, plus specialist agents for deployment, performance optimization and AI application design. It adds /vercel:deploy and /vercel:env commands, requires OpenCode 2, and connects to the Vercel MCP server through /mcps. Source
6. Vercel Observability Adds AI Traffic Insights
Vercel added AI traffic insights to Vercel Observability, showing how AI bots access a site. The view reports requests by verified AI bot, the pages most crawled for indexing or training, assistant retrievals triggered by user actions or searches, pages requested in Markdown format, and pages returning errors to AI bots. It is available under Observability > AI Traffic, and developers and agents can query the metrics with vercel metrics in the CLI, which requires Observability Plus. Source