Vercel — Cloud, Developer Tools product updates and releases, tracked on megachangelog.
Microsoft's MAI-Image-2.6 and MAI-Image-2.6-Flash models are now available on AI Gateway, enabling image generation and editing through a single API key without requiring separate provider accounts. Both models support photorealistic imagery, text rendering, targeted edits, and up to five reference images with output sizes up to 1536×1536.
Fish Audio's Drama 3 text-to-speech model is now available on AI Gateway with preview access and 80% discount for 30 days. The model enables natural-language voice direction for voiceovers, character dialogue, and audio storytelling with support for emotion, pacing, multi-speaker dialogue, and custom delivery tags.
Google's Gemini 4 Argon model is now available through Vercel's AI Gateway with support for coding, agentic work, and complex reasoning tasks. The model features a 1M token context window, multimodal input capabilities (text, image, PDF, video), tool calling, and configurable thinking levels for enhanced performance.
Repository checkout and Git LFS downloads are now faster on Vercel, reducing build preparation time. Average checkout time improved by 20%, with LFS-enabled projects seeing a 54% improvement and saving approximately 13 seconds per deployment.
Meta's Muse Wonder model is now accessible through Vercel's AI Gateway, supporting coding tasks, multi-step agent workflows, and computer use with a 1M-token context window. Users can access the model via a single API key without creating a separate Meta account and use it for text, image, PDF processing, tool calling, and structured outputs.
Vercel CLI now enables agents to discover, check availability, and initiate domain purchases through structured commands, with purchase confirmation remaining under user control via interactive terminal or dashboard approval.
Pro teams now default to a 30-day deployment retention period to balance keeping recent history available while reducing storage costs. You can configure a longer retention period in team settings if needed.
Pro teams will now be charged for Deployment Storage and Functions Storage at $0.10 per GB-month. The default deployment retention window changes to 30 days, with deployments older than 30 days deleted automatically unless you adjust your retention settings beforehand.
Rillet, an AI-native ERP platform, deployed AI agents on Vercel using the eve framework to automate internal workflows and customer requests, tripling their shipping rate within three months. The two-person engineering team now closes 800+ Linear tickets in four months and ships customer changes to production in as little as two hours, powered by standardized agents, Vercel Connect for integrations, and AI Gateway for model access.
Liquid AI's d1 decision model is now available on Vercel's AI Gateway, enabling structured classification and scoring without text generation. The model supports multiple APIs (Decision API, OpenAI-compatible, TypeSafe-compatible) and includes vision capabilities for image classification and inspection.
Microsoft Decision-1, a classification model for answering yes/no questions and categorizing text, is now available through Vercel's AI Gateway. The model supports choice classification, scoring, and boolean decisions with probability scores, accessible via AI SDK, OpenAI-compatible, and TypeSafe APIs using only an AI Gateway API key.
Vercel now allows you to skip sending client request bodies to Routing Middleware via the skipMiddlewareRequestBody configuration option, reducing middleware invocation costs and improving time to first byte for requests with large bodies.
Vercel Sandbox instances are now ready to run commands and access files immediately upon creation completion. The Sandbox.create() method now waits for snapshot restoration to finish before resolving, eliminating delays on the first command or file operation.
StepFun's Step 5 Preview model is now available on Vercel's AI Gateway, supporting text and image input with a 1M-token context window optimized for agentic coding, professional knowledge work, and financial analysis. The model can be accessed via the unified AI Gateway API for tracking usage, cost management, and performance optimization.
xAI's Grok Imagine Video 1.5 Lite model is now available on AI Gateway, enabling users to generate videos from text prompts and images with native audio support in multiple resolutions and aspect ratios. The model integrates seamlessly with AI Gateway's unified API for model management, cost tracking, and routing configuration.
Claude Haiku 5.5 is now available through Vercel's AI Gateway, featuring adaptive thinking with effort levels and zero data retention. The fastest Claude model is optimized for high-volume cost-sensitive tasks like summarization, classification, and live customer support.
OpenAI's Decisions API is now accessible through Vercel's AI Gateway with an OpenAI-compatible endpoint, enabling decision models to answer typed questions and return probabilities, choices, and scores for routing, triage, and guardrails use cases. Support is available across the OpenAI SDK, AI SDK, HTTP API, and CLI with the latest versions.
Vercel Flags now supports timestamp attributes for entities, allowing you to create time-based targeting rules. Use this feature to run limited-time campaigns, show content between specific dates, or target users based on registration date.
Microfrontends Routing charges are now waived for requests that the Web Application Firewall denies, challenges, or rate limits. This change applies automatically to all projects using microfrontends.
Glyph Cluster, a reasoning model for coding and long-context analysis, is now available as a stealth model on Vercel's AI Gateway for Pro and Enterprise plan teams with purchased AI Gateway credits at no cost during the stealth period. The model supports function calling, streams responses, and can be accessed via AI SDK, OpenAI-compatible APIs, and coding agents.
AI Gateway now supports escalating requests to a fallback model when the primary model's confidence falls below a threshold you set. Configure confidence conditions using providerOptions.gateway.models to automatically route low-confidence responses to secondary models, with support for Choice, Score, and Boolean questions.
Nano Banana 2.1, an improved image generation and editing model, is now available on Vercel's AI Gateway. The model offers better product recontextualization, mask- and ink-based editing, and factuality while maintaining Flash-tier latency and cost.
Mistral Large 4, an open-weight multimodal model supporting text and image understanding, is now accessible through AI Gateway with a single API key. You can use it via OpenAI-compatible Chat Completions, Responses, or Anthropic Messages APIs, or in coding agents without needing a separate Mistral account.
FLUX 3 Image from Black Forest Labs is now accessible through AI Gateway, enabling text-to-image generation and image editing with support for up to ten reference images and multiple resolution tiers from 768×768 to 4K. Users can access it via a single API key without needing a separate Black Forest Labs account.
AI Gateway now supports OpenAI's Ultrafast service tier for GPT-6 Astra, enabling faster output for interactive applications and rapid coding iterations. Ultrafast can be requested via AI SDK, Chat Completions API, or Responses API, with pricing at 6× the standard per-token rate.
Vercel released the AI SDK for Python with experimental support for Jev, a new AI model designed for making narrow classification decisions. Developers can now use the evaluate() API to build classifiers and decision-making tools without training custom models.
Rogo uses Vercel to deploy AI-generated code and manage production with agent swarms, achieving over 73,000 deployments monthly and zero manual incident triage through automated agent-based remediation.
Speed Insights will stop recording First Input Delay (FID) measurements on November 1st and will use Interaction to Next Paint (INP) as the responsiveness metric instead. No action is required on your side and existing FID data remains viewable.
Laya, an evaluation and decision model from Convai Innovations, is now available on Vercel's AI Gateway. It evaluates context against yes/no, choice, or scoring questions and returns structured answers with probabilities. The model is free to use through October 31, 2026 via Boundless.
Vercel's AI Gateway now supports Microsoft AI models including MAI-Voice-2.1, MAI-Voice-2.1-Flash for speech generation, and MAI-Transcribe-2 Streaming for live transcription. These models are billed at listed rates with no platform markup, offering users control over data and transparency with Microsoft's zero data retention policy.
Vercel Agent can now install private dependencies from npm and custom registries using team-shared environment variables. Credentials are securely stored outside the sandbox, and npm, pnpm, and classic Yarn authenticate the same way as in Vercel builds.
Browserbase Search and Fetch tools are now available through AI Gateway, allowing you to add web search and page retrieval capabilities to any model that supports tool calling. These helpers are included in AI SDK 7.0.116 and later.
Edge Requests has been renamed to CDN Requests across the platform, including usage dashboards, invoices, and notifications. Pricing, limits, and usage measurement remain unchanged, and saved queries continue to work.
Vercel CDN has changed its caching behavior to exclude responses with the Vary: Cookie header, as cookies often contain visitor-specific values that create cache entries with poor reuse rates. Users can identify affected responses via the x-vercel-cache header showing MISS and vary_key_denied:cookie in Runtime Logs, and can restore caching by removing Cookie from the Vary header or using Cache-Control: private if the response depends on cookies.
Vercel Sandbox now integrates with Secure Compute, allowing sandboxes to connect to dedicated networks with static IP addresses and access private AWS VPC resources through VPC peering. Users can attach sandboxes to networks via code or CLI, and existing sandboxes can update their network assignments.
Voyage AI's Rerank 3 and Rerank 3 Lite models are now available through AI Gateway for reordering search results by relevance. Rerank 3 prioritizes accuracy while Rerank 3 Lite optimizes for latency and cost, both supporting up to 32K tokens per query-document pair.
Ling 3.1 Flash, a hybrid reasoning model with 560B parameters optimized for coding and multi-step analysis, is now available on Vercel's AI Gateway and free to use through October 13, 2026. The model offers a 262K-token context window and is designed for agents using tools and workflows with long documents and extended task histories.
Users can now submit their own services to the Vercel Connect connectors directory instead of waiting for Vercel to build a connector. The submission process involves describing your service, adding OAuth or API key connection methods, verifying OAuth methods, and sending for review.
The Vercel CLI now supports searching trace spans directly from the terminal via the vercel traces search command, allowing developers to investigate errors, latency, and security issues without accessing the dashboard. The command returns up to 100 spans from the last hour by default and supports filtering by environment, service, span name, status, deployment ID, request ID, trace ID, span ID, and time range.
OpenAI's GPT-6.1 Sol model is now available on Vercel's AI Gateway with improved performance for coding, document analysis, and multi-step workflows. The model offers lower pricing than GPT-6 Astra and cheaper cached input, making it ideal for agents and applications that process complex tasks and reuse long context.
Claude Sonnet 5.5 from Anthropic is now available on Vercel's AI Gateway as anthropic/claude-sonnet-5.5. The new model improves coding task performance with better multi-step reasoning, enhanced document creation, and supports zero data retention across multiple API formats.
Domain name discovery and availability checking are now available without authentication through the Vercel CLI and Domains Registrar API, allowing users to search by keyword or check up to 200 domain names in a single request before signing in.
Fireworks' Ember-1, a reasoning model optimized for coding and agentic workflows, is now available on Vercel's AI Gateway. It generates approximately 40% fewer tokens than comparable models while maintaining quality, helping reduce costs and context overhead for agents making repeated calls.
You can now push container images from GitHub Actions to Vercel Container Registry using GitHub OIDC authentication, eliminating the need for long-lived registry credentials. The vercel/vcr-action/login action exchanges your workflow's OIDC token for a short-lived Vercel access token that is automatically revoked when the job ends.
Vercel's skills.sh registry reached one million agent skills in seven months with nearly 280 million installs. The report analyzes skill adoption patterns across categories, showing high demand for business operations and writing skills, while most installs concentrate on cross-industry and portable skills that apply broadly across jobs and industries.
The Vercel dashboard now includes a Redis database browser that lets you run Redis commands, browse and filter keys, and inspect values directly without needing external tools like redis-cli. It supports both Redis and Upstash for Redis integrations and includes features like transaction support, key grouping by prefix, and metadata inspection.
Pixel Canary, a new AI model optimized for coding and frontend development, is now available for free in stealth on Vercel's AI Gateway. The model achieves 90.3% success on Next.js evals and 96.8% with documentation, enabling developers to use it for app migrations, data fetching, and web interface implementation.
Vercel Sandbox now includes memory usage data accessible through the dashboard and CLI. You can view average, P75, and P95 memory metrics, track memory usage against the sandbox ceiling, and create custom queries and alerts using the memoryUsedBytes measure.
Vercel has consolidated its private and open-source bug bounty programs into a single public HackerOne program. The move expands researcher access to discover vulnerabilities across the entire Vercel platform while leveraging improved security processes and tooling to handle the increased volume of submissions.
Klaviyo launched a citizen developer program on Vercel that enabled 512 employees to deploy 356 full-stack internal apps in two weeks, with built-in security through SSO, Secure Compute, and identity provider integration requiring no builder configuration.
Vercel Connect adds support for TanStack AI agents to call OAuth-protected MCP servers with automatic token management and no credential storage required. The new connectMCPTransport subpath export integrates Connect-backed auth that provides fresh tokens before every request and handles consent challenges before the model executes.
Vercel Blob now allows unlimited store creation across all plans, removing previous limits of 100 (Hobby), 500 (Pro), and 1,000 (Enterprise). Store creation is billed as an advanced operation, enabling better data separation for production/staging environments, multi-tenant applications, and preview deployments.
Google's Gemini 3.8 Flash-Lite TTS and Gemini 3.8 Flash TTS models are now available on Vercel's AI Gateway, supporting 100+ languages with long-form narration, delivery control, and multi-speaker dialogue. Users get unified API access, usage tracking, and cost per-request visibility alongside other models.
Claude Opus 5.5 from Anthropic is now available via Vercel's AI Gateway with improved agentic coding, long-context reasoning, and a 1M-token context window. The model includes two breaking API changes: adaptive thinking (no longer allows disabling thinking or fixed budgets) and removal of forced tool use in favor of structured outputs.
Vercel's AI Gateway now supports OpenAI's GPT-6 Sol and Luna models, offering improved performance in coding, professional work, and factuality at lower cost than GPT-6 Astra. Sol is optimized for complex workflows requiring quality iteration, while Luna is the cost-effective option for high-volume agentic tasks.
Drives provide persistent storage that can be mounted as directories in Vercel Sandbox and reused across multiple sandbox instances. Each Drive supports up to 16 TiB of configurable storage with read/write pricing, enabling users to preserve agent workspaces, datasets, and models across sandbox runs.
Vercel Connect now includes a managed connector for Microsoft Teams, allowing organizations to create a Teams bot that their apps and agents can run. The bot can be mentioned in channels or messaged directly, and Vercel handles Entra app registration and Azure Bot resource management without requiring manual client secret storage.
Deployments now display their billable duration and CPU minutes in the dashboard, REST API, and CLI, helping users understand how each build contributes to their usage. Billable duration is the combined build and post-build time rounded up to the next whole minute, and CPU minutes are that figure multiplied by the machine's vCPUs.
Grok 4.7 from xAI is now available on Vercel's AI Gateway with a 500K token context window, configurable reasoning levels, and 40% off pricing through September 27. The model works with AI SDK, OpenAI-compatible APIs, and coding agents like fx and eve.
Xiaomi's MiMo V2.6 model family (Pro, Flash, and Pro UltraSpeed) is now available on Vercel's AI Gateway. These multimodal models support text, image, audio, and video understanding with 1M token context, structured outputs, and tool use capabilities for coding and agent workflows.
AI Gateway now supports calling Jev, a probabilistic decision model from TypeSafe AI, through TypeSafe clients, an HTTP API, or the AI SDK. All requests are billed through AI Gateway and appear in unified usage and observability, with support for boolean, choice, and score evaluation types.
Spend Management is now available at no additional cost to Enterprise teams on Flexible Commitment plans. Teams can set budgets per billing cycle and receive email notifications, trigger webhooks, or pause production deployments when metered usage approaches the budget threshold.
mcp-handler now has experimental support for WebMCP, a web standard for exposing tools to in-browser agents. Add a script tag to your site to make your MCP tools available to browser agents, with authentication handled transparently for signed-in users.
v0 now installs private packages from npm and custom registries using credentials stored as shared environment variables on Vercel, making it easier for teams to build with their existing design systems and internal packages directly in v0.
Jev, a specialized probabilistic decision model from TypeSafe AI, reached 13% of paid teams within 24 hours of launch on AI Gateway, doubling the adoption of any previous model. The model is optimized for structured decision-making in software with typed output and was reported to be up to 194x faster and 445x cheaper than general-purpose language models.
GLM 5.3 FlashX, a high-speed serving option for Z.ai's multimodal coding model, is now available on AI Gateway, delivering inference at ~200 tokens per second. This enables faster streamed responses for coding agents, tool loops, and interactive applications.
A remote code execution vulnerability was discovered in libheif, an AVIF image decoder used upstream by Next.js image optimization. Vercel coordinated with Hacktron and open source maintainers to responsibly disclose and fix the issue, deploying platform-wide mitigations and releasing patched versions across the dependency chain.
Vercel CLI now supports deploying static artifacts (HTML, HTM, and Markdown files) to live URLs in under one second by skipping the build step for eligible deployments. Users can share prototypes, publish reports, or preview agent-generated pages with a single command.
You can now opt into Turbo build machines on a per-deployment basis via Git commit message, CLI flag, or REST API, allowing you to temporarily increase build resources without modifying project settings.
You can now run Harbor evaluations, including Terminal-Bench and other benchmarks like SWE-bench and OSWorld, on Vercel Sandbox. Each trial runs in an isolated Firecracker microVM with network policy enforcement and optional credential injection, allowing you to parallelize benchmarking beyond local machine capabilities and test multiple models via AI Gateway.
The skills CLI now accepts Notion-hosted skills databases as an install source, allowing teams to author, review, and manage reusable agent skills directly in their Notion workspace without requiring a Git repository. Access and installation permissions follow Notion's built-in page sharing model.
Vercel's monthly AI Gateway Production Index shows open-weight models processed 56% of tokens in August 2026, their first month capturing majority volume. The data reveals shifting competitive dynamics as Anthropic's Opus 5 gained share when Fable 5 lost demand, while OpenAI's new GPT-6 Astra captured one-third of OpenAI spend within two days of launch, outpacing Anthropic's Fable 5.1.
GPT-Live 1, a full-duplex voice model from OpenAI, is now available on Vercel's AI Gateway. It can listen and speak simultaneously without turn detection, and supports client delegation to run background text models while maintaining the conversation.
Native Marketplace resources can now be connected to custom environments in addition to production, preview, and development. This lets you scope environment variables created by connections to specific custom environments via the dashboard, CLI, or REST API.
Hobby projects now keep only their 3 most recent production deployments and 3 most recent deployments of any type, with older deployments deleted immediately if a team exceeds the 10GB storage limit. This change helps dormant projects stop consuming storage needed by active projects.
Builds using Secure Compute or Static IPs now start 64% faster by using prewarmed containers with network configuration attached at build start, reducing average startup time from 6.7 to 2.4 seconds with no configuration changes required.
Mem0 is now available as a native integration on the Vercel Marketplace, providing AI agents and apps with long-term memory capabilities for user preferences, facts, and context across sessions. Setup is simplified with integrated billing, automatic provisioning of a scoped Mem0 project and API key, and no separate account management required.
TypeSafe AI's Jev, a probabilistic decision model for software, is now available on AI Gateway. Jev evaluates questions in parallel and returns typed answers with probabilities directly, enabling faster and cheaper decision-making compared to traditional language models for structured reasoning tasks.
Is Agentic audit reports now let you filter and view checks across four site types—Docs & content, Business, App, or Commerce—each highlighting checks relevant to that site category. You can set your default view with a meta tag, and your score remains consistent across all views.
Delphi rebuilt its Python backend on Vercel and now ships 100+ times a day with feature flags, eliminating heavyweight staging deployments and enabling faster iteration. New engineers can reach production in minutes instead of a day, and the entire team—including product and design—can deploy changes as everyday work.
Google's Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking models are now available through Vercel's AI Gateway, supporting real-time spoken interactions via the realtime API with features like multi-language support and parallel reasoning.
The AI SDK harness layer now authenticates through native subscriptions where supported by the underlying harness. Credentials remain on the host while the harness receives placeholder credentials, enabling seamless agent switching without code changes across adapters like Claude Code, Cline, and GitHub Copilot.
Vercel Sandbox now provides 64 GB of storage, increased from 32 GB, across all sandbox types including those created from Vercel Managed Images or custom images. This additional capacity supports larger repositories, dependencies, build artifacts, and disk-based data workloads.
Tailscale built Aperture, a customer-facing AI model router on Vercel's AI Gateway and Sandbox, enabling companies to control model access through network identity while ensuring zero data retention and safe agent execution without building infrastructure in-house.
Featured, a PR co-pilot platform, migrated from AWS Elastic Beanstalk to Vercel, enabling their 3-engineer team to handle 100,000+ monthly media pitches across 374 sites. The move simplified infrastructure management by consolidating single-click deploys, unified AI model routing through AI SDK and AI Gateway across 17 models, and enabled rapid feature development using Workflow SDK and Vercel's eve agent framework.
Pro and Enterprise teams can now restrict who creates and manages connectors using a new Connector Manager permission. Once enabled, only Owners and users with the Connector Manager role can manage credentials and external service access.
The AI SDK harness layer now supports GitHub Copilot through an official adapter, enabling developers to switch between different coding agents without changing application code. The harness layer uses the Agent Client Protocol (ACP) to connect multiple coding agents including Claude Code, Cline, Cursor, and others.
FastAPI frontends and static files are now automatically promoted to the Vercel CDN at build time, serving requests directly from the CDN without invoking your Function. Route precedence and runtime access are preserved, with configuration options to exclude directories or override CDN behavior.
Tako Search, a model-agnostic search tool that combines curated knowledge with live web data and source citations, is now available for free on AI Gateway through September 30th. Developers can integrate it via the AI SDK using gateway.tools.takoSearch() in text generation requests.
Vercel optimized CDN metadata lookup by grouping path metadata into shards instead of storing individual cache entries per path. This reduced P99 latency by 91% and sped up deployments, especially for large sites with hundreds of thousands of paths.
Vercel Sandbox is now available across all 20 compute regions instead of four, enabling lower latency and better regional control. Teams can set default regions, configure failover options, and ensure compliance with data residency requirements.
Vercel now integrates with OpenAI Agents API to enable long-running, tool-using agent workflows hosted on Vercel's infrastructure. Developers can leverage OpenAI's managed agent loop combined with Vercel Sandbox for isolated code execution and Vercel Queues for webhook processing, creating scalable agent applications without maintaining virtual machines.
Vercel Authentication and Deployment Protection are now available at no cost on all plans, allowing users to protect production and preview deployments without the $150/month add-on. Deployment Protection Exceptions also let you selectively make specific preview domains public.
Pro teams can now purchase Password Protection on a per-project basis for $20 per month instead of team-wide, allowing visitors to enter a password to view project deployments. This provides more granular control and cost management for teams that only need protection on specific projects.
v0 now supports one-click integrations with providers like Resend, Clerk, MongoDB Atlas, Algolia, and Amazon OpenSearch. Users can prompt v0 to build features, and when a provider is needed, v0 renders a connect card in chat to automatically handle environment variables, configuration, and provider-specific agent skills.
Eve agents can now retain context across sessions using persistent memory organized into slots with configurable providers and scopes. Memory is automatically retrieved before each turn and added to the model's context, with support for file storage, Supermemory, Upstash, and custom providers.
DeepSeek V4.1 Flash is now available on Vercel's AI Gateway with native image understanding, supporting text and images in the same request, a 1 million token context window, and responses up to 384,000 tokens. The model includes reasoning, tool use, and prompt caching capabilities with optimized architecture for reduced computation.
You can now read and search the Vercel changelog feed directly from your terminal using the vercel changelog command, allowing you and your agents to discover new products, features, and updates relevant to your projects.
The deployment step is now approximately 10% faster by combining separate routing metadata files into a single manifest. This optimization applies automatically to all builds without requiring any changes.
Vercel Sandbox now resolves public domains from nearest regional replicas instead of a centralized store, reducing median domain lookup latency from 62ms to 3.4ms and improving performance by up to 146x in remote regions like Cape Town.
97 more changes from Vercel. Sign up to read the whole changelog.
Sign up freeGitHub or email — no card needed.