Ce que Vercel a publié en septembre 2026

Vercel compares GPT-6.1 Sol vs. GPT-6 Luna for AI Gateway users

Vercel published a guide distinguishing GPT-6.1 Sol and GPT-6 Luna, two new OpenAI models accessible through Vercel’s AI Gateway. Sol targets complex coding and professional workflows with higher toke…

Vercel explains how to integrate OpenAI's GPT-6 models into applications

Vercel published a guide detailing OpenAI’s GPT-6 family (Astra, Sol, Luna), their capabilities, pricing, and how developers can integrate them into applications. The post emphasizes Vercel’s role in …

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

Vercel published a detailed guide comparing OpenAI’s Decisions API and TypeSafe AI’s Jev, highlighting differences in input support, answer types, and output formats like probabilities and rubric scor…

Vercel enables GPT-6.1 Sol and GPT-6 Luna workflows with AI Gateway

Vercel introduced new workflow patterns for GPT-6.1 Sol and GPT-6 Luna models, demonstrating support workflows, document assistants, and coding agents via the Vercel AI Gateway. The post provides code…

Vercel explains OpenAI's Decisions API for workflow routing and classification

Vercel published a detailed guide on OpenAI's Decisions API, which uses GPT-6 Luna to answer questions with predefined choices for routing, classification, or agent actions. The API is in limited prev…

Vercel adds Voyage Rerank 3 and Rerank 3 Lite to AI Gateway

Vercel integrated Voyage AI's Rerank 3 and Rerank 3 Lite models into its AI Gateway, improving search relevance for long documents and code. Rerank 3 prioritizes accuracy, while Rerank 3 Lite optimize…

Vercel AI Gateway adds Browserbase Search and Fetch tools

Vercel’s AI Gateway now supports Browserbase Search and Fetch tools, enabling models to perform web searches and retrieve page contents via a unified API. The tools are accessible in AI SDK 7.0.116+ a…

Vercel showcases 5 practical ways to use ChatGPT Pro 500

Vercel published a guide demonstrating five practical workflows for ChatGPT Pro 500, including building prototypes in v0, debugging with Codex, running maintenance tasks, producing decision briefs, an…

Vercel explains OpenAI’s ChatGPT Pro 500 pricing and Astra Ultrafast for v0

Vercel published a guide detailing OpenAI’s new $500/month ChatGPT Pro 500 tier, which includes 25x the usage allowance of ChatGPT Plus and access to Astra Ultrafast in Codex and ChatGPT Work. The pos…

Vercel renames Edge Requests to CDN Requests in usage and billing systems

Vercel rebranded the 'Edge Requests' usage metric to 'CDN Requests' to better reflect that these requests are served by its CDN. The change is purely semantic—pricing, measurement, and limits remain u…

Vercel CDN stops caching responses with Vary: Cookie

Vercel’s CDN no longer caches origin responses when the Vary header includes Cookie, as cookies often create unique cache entries. Responses are still served but marked as MISS in the x-vercel-cache h…

Vercel AI Gateway adds InclusionAI's Ling 3.1 Flash model

Vercel integrated InclusionAI's Ling 3.1 Flash model into its AI Gateway, offering a 560B-parameter hybrid reasoning model with a 262K-token context window. The model is free to use through October 13…

Vercel Sandbox adds Secure Compute support for private network access

Vercel Sandbox now supports Secure Compute, enabling sandboxes to connect to a team’s dedicated network with static IP egress and AWS VPC peering for private resource access. Enterprise teams can atta…

Vercel Agent gains private npm package installation support

Vercel Agent can now install private dependencies from npm and custom registries using shared environment variables like NPM_TOKEN and NPM_RC. The feature leverages existing authentication methods use…

Vercel CLI adds `vercel traces search` for terminal-based span analysis

Vercel introduced `vercel traces search`, a new CLI command enabling developers to query and filter trace spans directly from the terminal. The command supports filtering by environment, service, stat…

Vercel Connect now accepts third-party service submissions

Vercel opened its Connectors directory to third-party submissions, allowing developers to propose new integrations without waiting for Vercel to build them. Submissions require service description, co…

Vercel adds OpenAI’s GPT-6.1 Sol to AI Gateway with lower pricing

Vercel’s AI Gateway now supports OpenAI’s GPT-6.1 Sol, improving coding, document analysis, and multi-step workflows. The model offers lower standard and cached input pricing than GPT-6 Astra and inte…

Vercel enables unauthenticated domain search via CLI and API

Vercel introduced public domain search functionality, allowing users to check availability and pricing without signing in or using an access token. The feature is available through the Vercel CLI and …

Vercel adds Anthropic's Claude Sonnet 5.5 to AI Gateway

Vercel integrated Anthropic's Claude Sonnet 5.5 into its AI Gateway, improving coding workflows, document generation, and multi-step task execution. The model supports Zero Data Retention and is acces…

Vercel adds Ember-1 from Fireworks to AI Gateway

Vercel integrated Ember-1, a research preview reasoning model from Fireworks, into its AI Gateway. Ember-1 offers 40% fewer generated tokens than Kimi K3 at similar quality, supports 1M-token context …

Vercel adds Redis CLI and key browser to Marketplace database tools

Vercel’s Marketplace database browser now supports Redis and Upstash for Redis, allowing team owners to run commands, inspect keys, and view metadata directly in the Vercel dashboard. The new CLI and …

Vercel launches skills.sh registry for AI agent skills, hits 1M skills in 7 months

Vercel’s skills.sh registry, launched in 2026, reached one million agent skills and nearly 280 million installs in just seven months, outpacing GitHub repositories, App Store apps, and npm packages in…

Vercel adds Pixel Canary to AI Gateway as free stealth model for coding agents

Vercel introduced Pixel Canary, a new AI model optimized for coding and frontend development, now available for free in stealth mode on Vercel AI Gateway. It matches top performance on Next.js evals (…

Vercel Sandbox gains memory observability with P75/P95 metrics

Vercel added memory usage tracking to Sandbox observability, including average, P75, and P95 metrics in the dashboard and CLI. The feature surfaces memory limits via scaled charts and a 85% reference …

Vercel launches GitHub Actions integration for Vercel Container Registry

Vercel introduced a new GitHub Actions integration that allows pushing container images to Vercel Container Registry (VCR) using short-lived OIDC tokens instead of long-lived credentials. The `vercel/…

Vercel launches public bug bounty program to harden platform security

Vercel transitioned its private and OSS bug bounty programs into a single public program on HackerOne, covering all Vercel products and open-source projects. The move aims to leverage AI-driven resear…

Vercel Connect integrates with TanStack AI for OAuth-protected MCP servers

Vercel Connect now supports TanStack AI agents by enabling OAuth-protected MCP server calls without credential storage. The new `@vercel/connect/tanstack-ai` subpath provides `connectMCPTransport` for…

Vercel customer story: Klaviyo ships 356 internal apps in two weeks using Vercel

Klaviyo built a citizen developer program on Vercel, enabling 512 employees to deploy 356 apps—196 full-stack—within two weeks. Every app ships SSO-gated, private by default, and connects to Klaviyo’s…

Vercel removes Blob store limits, introduces per-store billing

Vercel eliminated store limits on Vercel Blob, allowing unlimited store creation across all plans. Stores are now billed under Blob Advanced Operations ($5/million on Pro, free on Hobby within monthly…

Vercel adds Google's Gemini 3.8 TTS models to AI Gateway

Vercel integrated Google's Gemini 3.8 Flash-Lite TTS and Flash TTS models into its AI Gateway, enabling developers to generate speech in over 100 languages with advanced controls like tone, pacing, an…

Vercel adds Anthropic’s Claude Opus 5.5 to AI Gateway with performance and cost improvements

Vercel integrated Anthropic’s Claude Opus 5.5 into its AI Gateway, offering ~30% faster and ~40% cheaper performance than Opus 5 while excelling in agentic coding and long-running tasks. The model int…

Vercel AI Gateway adds OpenAI GPT-6 Sol and Luna models

Vercel’s AI Gateway now supports OpenAI’s GPT-6 Sol and Luna models, offering higher quality and lower cost alternatives to GPT-6 Astra. Sol targets complex professional workflows and coding, while Lu…

Vercel Sandbox Drives enter public beta with persistent storage

Vercel launched Drives for Vercel Sandbox in public beta, enabling persistent storage across sandbox instances. Users can mount Drives to preserve workspaces, datasets, or models, with read-write acce…

Vercel adds Grok 4.7 to AI Gateway with 40% discount and expanded reasoning levels

Vercel now supports Grok 4.7 from SpaceXAI on its AI Gateway with a 40% discount through September 27, 2026. The model features a 500K token context window and four reasoning levels (low to xhigh), en…

Vercel AI Gateway adds Xiaomi MiMo V2.6 models with 1M token context

Vercel’s AI Gateway now supports Xiaomi’s MiMo V2.6 family—Pro, Flash, and Pro UltraSpeed—offering 1M token context, multimodal understanding, and structured outputs. The models target complex coding,…

Vercel AI Gateway adds TypeSafe client and HTTP API support for Jev

Vercel’s AI Gateway now supports calling the Jev probabilistic decision model through TypeSafe clients, a new HTTP API, or the existing AI SDK. Existing TypeSafe clients can point to AI Gateway withou…

Vercel adds billable duration and CPU minutes to deployments

Vercel now exposes billable duration (build + post-build time) and CPU minutes in the dashboard, CLI (`vc inspect`), and REST API for deployments. Billable duration is rounded up to the next whole min…

Vercel Connect adds managed Microsoft Teams connector

Vercel launched a managed connector for Microsoft Teams in Vercel Connect, enabling organizations to deploy Teams bots that respond to @mentions or direct messages. The connector handles Azure Entra a…

Vercel mitigates libheif RCE vulnerability in Next.js image optimization

Vercel disclosed and mitigated a remote code execution vulnerability in Next.js image optimization caused by a flaw in the upstream libheif AVIF decoder. Vercel deployed platform-wide mitigations and …

Vercel AI Gateway adds Jev model from TypeSafe AI with record adoption

Vercel integrated Jev, a probabilistic decision model from TypeSafe AI, into its AI Gateway, achieving the fastest adoption in gateway history—used by 13% of paid teams within 24 hours, outperforming …

Vercel mcp-handler adds experimental WebMCP support for browser agents

Vercel’s mcp-handler now supports WebMCP, a proposed web standard for exposing tools to in-browser agents. A single script tag enables existing MCP tools to work in browsers, with tools opt-in via an …

Vercel expands Spend Management to Enterprise Flexible Commitment plans

Vercel extended its Spend Management feature to Enterprise Flexible Commitment plans at no extra cost, allowing teams to set budgets, receive notifications, trigger webhooks, or pause deployments when…

Vercel v0 now reads npm credentials from shared environment variables

Vercel’s v0 now supports installing private npm packages using credentials stored as shared environment variables, including NPM_TOKEN for npmjs.org and NPM_RC for custom registries. Credentials remai…

Vercel AI Gateway adds Z.ai's GLM 5.3 FlashX for coding agents

Vercel AI Gateway now supports Z.ai's GLM 5.3 FlashX, a high-speed multimodal coding model delivering ~200 tokens/second for faster streamed responses. The model is integrated with Vercel's AI Gateway…

Vercel enables Harbor evals on Vercel Sandbox with AI Gateway integration

Vercel now supports running Harbor evals—including Terminal-Bench and other benchmarks—directly on Vercel Sandbox. Each trial executes in an isolated Firecracker microVM with enforced network policies…

Vercel Marketplace integrations now support custom environments

Vercel expanded Marketplace integrations to support custom environments beyond production, preview, and development, allowing resource connections to target specific environments like staging. The fea…

Vercel skills CLI adds Notion-hosted agent skills support

Vercel's skills CLI v1.7.0 now supports installing agent skills directly from Notion databases, allowing teams to manage skills in Notion without Git repositories. Skills are installed via Notion page…

Vercel CLI now supports sub-second artifact deployments

Vercel CLI now enables sub-second deployments of static artifacts (HTML/Markdown files ≤5MB) via `vercel deploy`, skipping the build step. The feature supports up to 10 files and requires CLI version …

Vercel AI Gateway Index: Open-weight models surpass closed-weight in token volume

Vercel's AI Gateway Index for September 2026 reveals open-weight models processed 56% of gateway tokens in August, surpassing closed-weight models for the first time. Token costs fell 23.2% in August,…

Vercel adds GPT-Live 1 to AI Gateway with full-duplex voice and client delegation

Vercel’s AI Gateway now supports OpenAI’s GPT-Live 1, a full-duplex voice model enabling real-time interruptions and simultaneous listening/speaking. It also introduces client delegation, allowing app…

Vercel enables Turbo build machines per deployment

Vercel introduced the ability to enable Turbo build machines on individual deployments without altering project settings. Users can opt in via Git commit messages, Vercel CLI, or REST API, providing t…

Is Agentic adds site-type tailored audit views for Vercel-hosted sites

Is Agentic now categorizes audit reports into four site types—Docs & content, Business, App, or Commerce—with tailored checks highlighted per type. Scores remain comparable across views, and a meta ta…

Vercel Secure Compute and Static IP builds start 64% faster

Vercel reduced build start times for Secure Compute and Static IP deployments by 64%, from 6.7 seconds to 2.4 seconds, by prewarming build containers with attached network configurations. The change i…

Mem0 joins Vercel Marketplace with native integration and billing

Vercel added Mem0 to its Marketplace, enabling AI agents and apps to retain long-term memory across sessions. The integration includes automatic provisioning, environment variables, and integrated bil…

Vercel adds TypeSafe AI's Jev to AI Gateway for probabilistic decision-making

Vercel integrated TypeSafe AI's Jev model into its AI Gateway, enabling parallel evaluation of typed decisions (Choice, Score, Boolean) with up to 193.6x speed and 444.6x cost savings over LLMs. The m…

Vercel Hobby projects reduce deployment retention to free up storage

Vercel shortened deployment retention for Hobby projects to free up storage, keeping only the 3 most recent production deployments and 3 most recent deployments overall. Old deployments are now delete…

Vercel adds Google's Gemini 3.8 Live models to AI Gateway

Vercel now offers Google's Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking models on its AI Gateway, enabling real-time spoken interactions for voice assistants and conversational apps. Both mod…

Delphi case study: 100+ daily deploys on Vercel’s Python backend

Delphi rebuilt its Python backend on Vercel, enabling 100+ daily production deploys with no dedicated infrastructure team. Workflows and Queues handled long-running tasks and background jobs, while AI…

Vercel AI SDK harness layer gains native subscription authentication

Vercel’s AI SDK harness layer now supports native subscription authentication for supported harnesses, enabling seamless credential resolution without code changes. The feature works across multiple A…

Pricing change detected for Vercel

Pricing updated for Vercel: - Hobby: allowance changed from Limited usage caps to 100 GB bandwidth/month, 10 GB storage/month, 1K ISR reads/month, 1 environment, 1 project - Pro: allowance changed fro…

Featured scales PR media pitches to 100K/month using Vercel’s AI and workflow tools

Featured, a PR co-pilot platform, migrated 374 sites from AWS Elastic Beanstalk to Vercel and now handles 100,000+ media pitches monthly with a 3-person team. The move leveraged Vercel’s AI SDK, AI Ga…

Tailscale builds Aperture model router using Vercel AI Gateway and Sandbox

Tailscale leveraged Vercel’s AI Gateway and Sandbox to launch Aperture, a customer-facing model router that controls AI model access via tailnet identity. The integration enabled rapid deployment, zer…

Vercel Sandbox storage doubled to 64 GB

Vercel increased storage for all Vercel Sandbox environments from 32 GB to 64 GB, effective immediately. The change applies to sandboxes created from managed or custom images, including those using th…

Vercel adds granular connector management permissions in Vercel Connect

Vercel introduced role-based access control for Vercel Connect connectors, restricting creation and management to Owners and users with the Connector Manager extended permission. The feature is availa…

Vercel adds GitHub Copilot harness adapter to AI SDK

Vercel introduced an official `@ai-sdk/harness-github-copilot` adapter, enabling GitHub Copilot to work within its AI SDK harness layer. Developers can now switch between multiple coding agents (inclu…

Vercel integrates OpenAI Agents API for long-running agent deployments

Vercel now supports deploying long-running, tool-using agents via OpenAI's Agents API, with Vercel managing hosting and Sandbox environments for code execution and file access. The integration leverag…

Vercel makes Tako Search free on AI Gateway through September 30th

Vercel announced Tako Search, a model-agnostic search tool, is free to use on AI Gateway until September 30th. It integrates with any AI model via the AI SDK, combining curated knowledge graphs with l…

Vercel FastAPI frontends and static files now served from CDN at build time

Vercel now promotes FastAPI frontends and static files to its CDN at build time, serving requests directly from the CDN without invoking Vercel Functions. This applies to paths using `app.frontend()` …

Vercel reduces CDN metadata lookup latency by 91% with sharded routing

Vercel cut CDN metadata lookup latency by 91% (P99) and improved deployment speed by replacing per-path metadata caching with sharded, indexed JSONL files. The new system groups paths into ~200KB shar…

Vercel Sandbox now runs in all 20 compute regions

Vercel Sandbox expanded from four to all 20 compute regions, reducing latency for workloads by running closer to data sources and enabling regional failover for Pro and Enterprise teams. Default regio…

Pricing updated for Vercel: - Hobby: now includes Limited usage caps - Pro: promotion changed — $20 included credit to use across resources - Pro: now includes Included credit + pay-as-you-go - Enterp

Pricing updated for Vercel: - Pro: promotion changed — Includes $20 monthly credit to use across resources - Enterprise: promotion changed — Contact sales team for custom pricing

Vercel Pro teams now pay $20/month per project for Password Protection

Vercel introduced per-project pricing for Password Protection on Pro plans at $20 per project per month, replacing the previous team-level purchase model. Teams can enable or disable the feature indiv…

Vercel makes deployment protection free on all plans

Vercel now allows all deployments, including production domains, to be protected via Vercel Authentication at no extra cost on Hobby, Pro, and Enterprise plans. Previously, this required a $150/month …

Vercel v0 adds one-click integrations for email, auth, search, and databases

Vercel’s v0 now supports one-click integrations with Resend, Amazon OpenSearch, MongoDB Atlas, Algolia, and Clerk, enabling developers to connect providers directly within prompts. The feature automat…

Vercel adds CLI command to read and search changelogs

Vercel introduced a new `vercel changelog` CLI command allowing users to read, search, and retrieve changelog entries directly from the terminal. The command supports filtering by keyword, adjusting r…

Vercel adds persistent memory for eve agents with multi-provider support

Vercel introduced persistent memory for eve agents, enabling context retention across sessions via configurable memory slots. The feature supports multiple providers, including a built-in file memory …

Vercel AI Gateway adds DeepSeek V4.1 Flash with vision support

Vercel’s AI Gateway now supports DeepSeek V4.1 Flash, a multimodal model with vision capabilities, 1M token context, and 384K token output. The integration enables text-and-image inputs in a single re…

Vercel Pro teams get Flat Rate CDN GA with fixed monthly billing

Vercel launched Flat Rate CDN as generally available for Pro teams, replacing usage-based billing with a fixed monthly fee. The plan includes 1M requests and 1TB of data transfer, with spike protectio…

Vercel deployment step now 10% faster with unified metadata manifest

Vercel reduced deployment time by 10% (up to 12 seconds for large apps) by consolidating routing metadata into a single manifest instead of separate files. The change is automatic and requires no user…

Vercel introduces Flat Rate CDN for Pro teams to eliminate surprise bills

Vercel launched Flat Rate CDN for Pro teams, replacing usage-based billing with a fixed monthly fee that covers CDN requests, data transfer, and observability events. The model includes built-in spike…

Vercel Sandbox routing now 18x faster globally with regional replicas

Vercel Sandbox public domains now route requests 18x faster globally by resolving domains from regional replicas instead of a centralized store. Median lookup latency dropped from 62ms to 3.4ms, with …

Vercel adds OpenAI GPT Image 2.5 Flare and Sunburst to AI Gateway

Vercel integrated OpenAI’s GPT Image 2.5 Flare and Sunburst models into its AI Gateway, offering faster generation (Flare) and precise control (Sunburst) for image editing and creation. Both models su…

Vercel launches Ling 3.0 Flash Sante on AI Gateway for free

Vercel made Ling 3.0 Flash Sante, a 124B-parameter Mixture-of-Experts model optimized for medical reasoning, available for free on AI Gateway through October 4. The model supports deep research, evide…

Vercel integrates GPT 6 Astra into AI Gateway for agentic workflows

Vercel added OpenAI’s GPT 6 Astra to its AI Gateway, enabling developers to use the model for long-running agentic tasks like software engineering, data analysis, and website building. Integration sup…

Pricing updated for Vercel: - Pro: allowance changed from 100 GB bandwidth/month, 10 GB storage/month, 10K requests/month + included credit to Included credit: $20/month - Enterprise: promotion change

Vercel adds Basic build machines to Pro and Enterprise plans

Vercel introduced Basic build machines (2 vCPUs, 8 GB RAM) for Pro and Enterprise plans, offering a cost-efficient alternative to Elastic build machines. New projects default to Elastic, but teams can…

Cursor Cloud Agents integrate with Vercel Sandbox for isolated execution

Cursor Cloud Agents now run in Vercel Sandbox via Cursor's Self-Hosted Machines APIs, replacing Cursor's hosted machines. Vercel Sandbox provides isolated Firecracker microVMs for each agent request, …

Vercel adds Meta's Muse Spark 1.3 to AI Gateway with contributor tier

Vercel integrated Meta’s Muse Spark 1.3 into its AI Gateway, offering it in standard and contributor pricing tiers. The model improves agent work and coding with a 1M token context window and supports…

Vercel adds Google’s Gemini 3.8 Flash to AI Gateway with 50% discount

Vercel’s AI Gateway now supports Google’s Gemini 3.8 Flash model, offering a 50% discount through December 31, 2026. The model features a 1M token context window, supports text, image, PDF, and video …

Vercel expands free domain offer to include .app and .dev for Pro teams

Vercel extended its free domain offer for Pro teams to include .app and .dev TLDs, bringing the total eligible domains to eight. The offer remains one domain per team, with automatic renewal at standa…

Vercel offers 50% off GLM-5.3 on AI Gateway through DigitalOcean

Vercel and DigitalOcean are offering a 50% discount on the GLM-5.3 model via Vercel’s AI Gateway until September 8, 2026. The promo applies only to requests routed through DigitalOcean and requires us…

Pricing updated for Vercel: - Hobby: allowance changed from 100 GB bandwidth / month, 10 GB storage / month, 100 GB data transfer / month, 10K function invocations / month, 10K edge requests / month t

Vercel adds Alibaba’s Qwen 3.8 Max 0902 to AI Gateway for coding agents

Vercel’s AI Gateway now supports Alibaba’s Qwen 3.8 Max 0902, a model optimized for coding on large projects, long-horizon tasks, and agent workflows. The integration enables developers to use the mod…

Vercel AI Gateway adds Anthropic’s Claude Fable 5.1 with model fallbacks

Vercel AI Gateway now supports Anthropic’s Claude Fable 5.1, optimized for long, multi-stage tasks like agentic coding and research. The update includes cybersecurity and biology safety classifiers, w…

Vercel launches Fluid compute: a unified system for builds, sandboxes, and functions

Vercel introduced Fluid, a single compute system replacing fragmented primitives for builds, sandboxes, and functions. Fluid assembles machines on-demand, supports custom environments via Fluid images…

Vercel adds AWS PrivateLink to Pro and Enterprise plans

Vercel now offers AWS PrivateLink for Pro and Enterprise teams, enabling private connectivity between Vercel Functions/builds and AWS-hosted services without public internet exposure. The feature is i…