Ce que Vercel a publié en 2026

Vercel updates AI SDK and AI Gateway with new decision-making APIs

Vercel introduced `experimental_decide` and `decisionModel()` in its AI SDK to replace deprecated `experimental_evaluate` and `evaluationModel()` functions. The update aligns with OpenAI's native Deci…

Vercel adds AI Gateway decision fallbacks for translation review workflows

Vercel introduced a new AI Gateway feature that uses decision fallbacks to flag UI copy changes requiring translation review. The system compares old and new source text to detect meaning changes (e.g…

Vercel compares OpenAI and Perplexity Decisions APIs for developers

Vercel published a detailed technical comparison of OpenAI’s and Perplexity’s Decisions APIs, highlighting differences in request formats, answer types, and deployment options. The guide includes code…

Vercel AI Gateway adds confidence-based decision fallbacks in beta

Vercel introduced confidence-based decision fallbacks in AI Gateway, allowing applications to escalate uncertain AI decisions when confidence falls below a set threshold. The feature supports Choice, …

Vercel expands Jev use cases for AI applications with decision-making workflows

Vercel published nine practical use cases for Jev, its TypeSafe AI System One model, demonstrating how it can route forms, prioritize tickets, review tool calls, select models, categorize documents, a…

Vercel adds AI-powered CSV column mapping with Jev and Luna fallbacks

Vercel introduced a new feature enabling AI-driven mapping of CSV column headers to application fields using AI Gateway decision fallbacks. The system suggests mappings via Jev, escalates to GPT-6 Lun…

Vercel demonstrates GitHub PR triage with OpenAI's Decisions API

Vercel published a guide and Node.js example showing how to use OpenAI's Decisions API to automatically triage GitHub pull requests by review needs, such as identifying compatibility changes or migrat…

Vercel AI Gateway adds Mistral Large 4 with multimodal support

Vercel AI Gateway now supports Mistral Large 4, an open-weight multimodal model combining text and image understanding, accessible via a single API key without requiring a separate Mistral account. Th…

Vercel adds Nano Banana 2.1 to AI Gateway with improved image editing

Vercel now supports Google’s Nano Banana 2.1 on AI Gateway, enhancing image generation and editing with product recontextualization, mask/ink-based edits, and improved factuality. The model delivers p…

Vercel’s AI Gateway now supports confidence-based decision fallbacks, allowing requests to escalate to a secondary model when the primary model’s confidence drops below a set threshold. Conditions can

Vercel publishes guide to integrating Perplexity's Decisions API

Vercel published a technical guide explaining how to integrate Perplexity’s Decisions API into applications, covering typed predictions, image support, and workflow design. The API evaluates text and …

Vercel compares TypeSafe AI's Jev and Perplexity's Decisions API for AI workflows

Vercel published a detailed technical comparison of TypeSafe AI's Jev and Perplexity's Decisions API, focusing on input support, question types, deployment options, and migration considerations. The g…

Vercel compares Jev, Laya, and Liquid d1 decision models for AI Gateway

Vercel published a detailed comparison of three decision models—Jev, Laya, and Liquid d1—available via AI Gateway and direct integrations. Laya uniquely offers self-hosting with downloadable checkpoin…

Vercel demonstrates Perplexity's Decisions API integration for image classification

Vercel published a technical guide showing how to use Perplexity's Decisions API to classify support screenshots by routing them to specialist queues based on image content and customer messages. The …

Vercel adds experimental Jev classifier API to Python AI SDK

Vercel introduced an experimental `evaluate()` API in its Python AI SDK to interact with the Jev model, a universal classifier designed for structured decision-making. The update enables Python develo…

Rogo ships 73K deployments/month with agent-written code in 5 minutes on Vercel

Rogo, an AI agent platform for financial institutions, deployed 73,000 times in a single month on Vercel, with agent-written code shipping to production in just 5 minutes. Their production AI agents u…

Vercel AI Gateway adds Microsoft AI models for speech and transcription

Vercel integrated Microsoft AI's latest audio models (MAI-Voice-2.1, MAI-Voice-2.1-Flash, and MAI-Transcribe-2 Streaming) into its AI Gateway platform. The models enable expressive multilingual speech…

Vercel Speed Insights deprecates First Input Delay on November 1st

Vercel's Speed Insights will stop recording First Input Delay (FID) measurements on November 1, 2026, as it transitions to Interaction to Next Paint (INP) as the responsiveness metric. No action is re…

Vercel enables ChatGPT plugin extensions with MCP Apps and Next.js starter

Vercel published guidance on building ChatGPT plugin extensions using MCP Apps, including sidebar apps, conversation panels, and file viewers. The company provides a Next.js starter template and hosti…

Vercel explains decision models and their role in AI workflows

Vercel published a guide defining decision models as AI systems that evaluate context against structured questions to return typed judgments (Choice, Score, Boolean) with probabilities. The post highl…