The latest AI news and updates of Vercel — Frontend cloud platform and creator of the AI SDK for building AI-powered web applications. Covering Vercel's latest product updates, launches, and analysis from the past 90 days.
v0Vercel1h agoVercel CEO Guillermo Rauch reports that internal teams now iterate as quickly on Zig, Go, and Rust as on TypeScript and Python. He asserts that agentic coding acts as a new compiler, translating intent into software and rendering language choice based on human convenience obsolete.
Vercel1h agoVercel Labs connects a 166,700-neuron simulated fly brain to agent-browser, the company's browser automation CLI. This integration provides the neural simulation with direct browser access. Chris Tate, the creator of agent-browser, demonstrated the connection, which gives the simulated neurons the ability to navigate web interfaces.
VercelSep 11Vercel added GitHub Copilot to the AI SDK harness layer via the new @ai-sdk/harness-github-copilot adapter. This integration connects Copilot to the unified HarnessAgent interface using the Agent Client Protocol. Applications now run GitHub Copilot alongside other supported coding agents without requiring changes to the underlying application code.
v0Sep 9v0 now supports native Marketplace integrations, including Resend, Amazon OpenSearch, MongoDB Atlas, Algolia, and Clerk, directly within the chat interface. When a prompt requires a provider, v0 renders a connect card for one-click setup, automatically configuring environment variables and loading provider-specific agent skills to ensure generated code follows recommended patterns.
VercelSep 9Vercel added OpenAI's GPT Image 2.5 to its AI Gateway, enabling image generation and editing through the AI SDK. The model, identified as openai/gpt-image-2.5-sunburst, supports text-to-image prompts and reference image editing. Access is available via the generateImage function, with a live playground for testing capabilities at the Vercel-powered imagen.sh site.
VercelSep 8Vercel Labs updated its agent-browser CLI to support 60fps video recording in version 0.37.0. Passing --fps 60 to the record command activates the higher frame rate. This improves capture quality for motion-heavy agentic workflows like WebGPU animations, and requires ffmpeg installed on the system.
VercelSep 5OpenAI’s Astra model now leads the DeepSecBench cybersecurity benchmark, completing tasks in 49 minutes compared to the previous leader’s four-hour runtime. While Astra’s cost per task is slightly higher than the prior benchmark leader, it remains 50% cheaper than Claude Opus 5 max.
v0Sep 5v0 now supports OpenAI’s GPT-6 Astra for building full-stack web applications. The frontier model, designed for complex agentic workflows, is available for selection within the v0 interface. The platform uses the model to design, iterate, and scale applications, providing access to its reasoning capabilities for web development tasks.
VercelSep 4Vercel added OpenAI's GPT-6 Astra to its AI Gateway, providing access via the openai/gpt-6-astra identifier. The model supports long-running agentic tasks including software engineering, computer use, and scientific research. It handles evolving instructions by incorporating new requirements without losing previous constraints and performs independent work while awaiting clarification.
VercelSep 3Vercel now hosts Cursor Cloud Agents within its Sandbox environment. Each agent request runs in an isolated Firecracker microVM, utilizing a scale-to-zero worker pool for efficient execution. Vercel Functions and Workflow provide a durable control plane to manage agent sessions, handle retries, and provision short-lived, user-scoped credentials for every task.
VercelSep 3Vercel and ora.ai analyzed over 1,000 agent runs and found that server-side rendering and accurate HTTP status codes are the primary drivers of website readability for AI agents. The study shows that adding metadata files like llms.txt does not compensate for content hidden behind JavaScript or blocked by 403 errors, emphasizing the importance of fundamental web accessibility.
VercelSep 1Vercel Labs added support for Anthropic’s Claude Fable 5.1 model to the fx coding agent. The model is available immediately within the command-line interface via the /model command. This update requires no manual upgrade to the local fx installation.
VercelSep 1Vercel launched Fluid Compute, a unified infrastructure system that dynamically assembles machines for builds, sandboxes, and functions on demand. The architecture uses Hive for hardware provisioning, Fluid images for environment management, and Vercel Drives for portable storage. This system runs workloads on a single compute layer, improving boot times and isolation while maintaining state across sessions.
v0Sep 1Vercel integrates its v0 app builder into Anthropic's Claude Design. The integration converts designs sent from Claude Design into full-stack applications, which then deploy directly to production.
VercelSep 1Vercel introduced design.md, a public file that encodes design guidance for AI agents to ensure generated pages maintain brand consistency. The system pairs this guidance with a public stylesheet and an evaluation loop, which uses human feedback to refine rules and catch mechanical failures. This approach helps agents avoid generic design patterns and align outputs with Vercel’s standards.
VercelSep 1Vercel's open-source fx coding agent is now available through the AI SDK HarnessAgent interface. The new @ai-sdk/harness-fx adapter lets you run fx via the unified HarnessAgent API, connecting over the Agent Client Protocol. This brings fx into consistent agent execution alongside other supported runtimes without requiring separate custom integrations.
VercelAug 31Vercel added per-user spending limits to AI Gateway, allowing teams to cap individual member usage alongside existing team, project, and API key budgets. These limits can be configured as a default for all members or customized per person. When a user reaches their allocated budget, the gateway rejects further requests until the limit resets or is increased.
VercelAug 31Vercel is offering a 50% discount on MiniMax H3 and H3 Max video generation models through its AI Gateway. The promotional pricing applies to all supported durations and aspect ratios through September 13, 2026. These models integrate with the AI SDK to support asynchronous video generation, enabling background processing for long-running rendering tasks.
VercelAug 30Guillermo Rauch speculates that JFrog's newly published CVE-2026-82329—a CVSS 9.8 Artifactory auth bypass—may be the vulnerability autonomous AI agents recently discovered and exploited. He argues defenses must match machine-speed attacks. Vercel provides the open-source deepsec security harness for automated vulnerability analysis.
VercelAug 29Guillermo Rauch describes the web evolving into two distinct extremes: ultra-fancy human experiences powered by WebGPU tools and agent-centric, machine-readable data. He predicts that agents will subsume utilitarian web content by generating just-in-time user interfaces. In this model, the agent functions as the new browser for accessing data and content.
VercelAug 28Vercel added Tencent Hunyuan's Hy4 Preview to AI Gateway. This open-source Mixture-of-Experts model features 770B total parameters, 49B active parameters per token, and a 1M-token context window. The integration supports unified API calls, spend tracking, and automatic fallbacks for long-horizon coding, document analysis, and scientific reasoning tasks.
VercelAug 28Vercel launched a one-minute setup flow for building eve agents. The process requires a Vercel account and involves adding a prompt, selecting models and MCP connections, and deploying. The resulting agent runs in production, ready for chat, and remains backed by a Git repository owned by the user.
VercelAug 28Vercel integrated Cursor into the AI SDK HarnessAgent interface via the new @ai-sdk/harness-cursor adapter. This update enables switching between coding agents without modifying application code. The harness layer now supports Cursor alongside Claude Code, Cline, Codex, and other agent runtimes by utilizing the Agent Client Protocol for unified execution.
VercelAug 28Vercel open-sourced vgpu, a minimal WebGPU library designed for AI agents to render and verify shaders. It runs in the browser or headless Node.js, supports reusable WGSL modules, and enables shader rendering within CPU sandboxes and CI tests. The library includes a CLI for documentation, shader validation, and MCP integration.
VercelAug 27Vercel launched Workflow SDK, an open-source framework that compiles durable execution graphs directly from plain TypeScript. It uses "use workflow" and "use step" directives to handle retries and state, while ad-hoc webhooks manage human-in-the-loop approvals. The SDK supports swappable backends like Postgres and Redis, pinning runs to specific deployments to prevent versioning errors in flight.
VercelAug 27Vercel added Inclusion AI's Ling 3.0 Flash Fin to AI Gateway, offering free access through September 25. The model features a 256K token context window and supports reasoning and function calling for financial research tasks. Using the identifier with a -free suffix ensures the model stops serving automatically when the offer period ends.
VercelAug 27Vercel's Chat SDK now runs Claude Managed Agents, supporting agent deployment across Slack, WhatsApp, Discord, Teams, and 30+ other platforms. The integration features token-by-token streaming and a live activity feed, while storing conversation state within the agent session to eliminate server-side database requirements. A one-click template deploys a research agent directly to these messaging services.
VercelAug 27Vercel added Meta’s Muse Image model to AI Gateway, enabling text-to-image generation and image editing through the AI SDK. The model supports reference images for style blending and editing instructions via the generateImage function. Vercel provides access at provider pricing with no platform markup or inference fees, including support for Bring Your Own Key requests.
VercelAug 27Vercel added Google's Gemini 3.5 Transcribe to AI Gateway, supporting both live WebSocket streaming and complete file recordings. The model features automatic detection for 85+ languages, including mid-speech language switching and custom vocabulary support. Integration occurs through the Vercel AI SDK using dedicated identifiers for live and batch transcription tasks.
VercelAug 26Vercel added Alibaba's Qwen 3.8 Flash to its AI Gateway, supporting text and image inputs with a 1-million-token context window. The model returns up to 65k tokens per response and targets coding, tool use, and multi-step agent workflows. Access is available via the alibaba/qwen3.8-flash identifier with no platform markup on inference costs.
VercelAug 26Vercel added Z.ai's GLM-5.3-Flash to AI Gateway, providing access to the multimodal model via the zai/glm-5.3-flash identifier. The model supports text and vision inputs with a 1-million-token context window. Z.ai is offering a 50% discount on inference costs through September 9th at 9:00 am PT.
VercelAug 25Vercel launched the Run SDK, a secure alternative to eval for executing agent-generated code. It runs TypeScript and JavaScript in a hardened QuickJS sandbox, exposing only approved host functions to the application. The SDK supports pausing execution for human-in-the-loop approval or authentication, resuming later without repeating completed work. It runs on Node.js 22.13+ and Bun.
VercelAug 25Vercel is offering free access to MiniMax M3 and M2.7 models on AI Gateway through September 6, 2026. Requests served via GMI Cloud using the dedicated free model identifiers or provider pinning incur no inference costs. The promotion applies to all AI Gateway API integrations and supports setup within coding agents.
VercelAug 25Vercel Connect adds a managed Linq connector, providing AI agents with a phone number for iMessage, RCS, and SMS messaging. The connector handles account provisioning and credential management through the Vercel CLI. It supports read receipts, typing indicators, and conversation continuity, enabling agents to send and receive messages across these protocols without manual credential configuration.
VercelAug 25Vercel added asynchronous video generation to AI Gateway, allowing jobs to run in the background instead of holding open long HTTP requests. Developers can now use webhooks, polling, or a start-and-status pattern to retrieve results. This update requires an SDK upgrade and enables monitoring of pending video generation tasks directly through the AI Gateway logs page.
VercelAug 25Vercel Connect is now generally available, replacing long-lived provider secrets with short-lived, scoped tokens for agents and apps. The system supports over 100 services, including Slack, GitHub, and Linear. It adds governance features like RBAC, audit trails, and token observability, ensuring agents access only the specific resources required for each task without storing permanent credentials.
VercelAug 25Vercel Labs updated its emulate library to support GitHub App emulation in TypeScript. The tool generates App keys, handles installation authentication, and provides stateful repository APIs. This allows building on GitHub integrations locally, testing in CI, or running sandboxed agents with isolated GitHub environments without requiring real third-party credentials or external network access.
VercelAug 22Vercel is running a two-week public hacker challenge on HackerOne, offering up to $1,000,000 in total bounties for sandbox escapes. Researchers can earn up to $50,000 per report for demonstrating vulnerabilities that bypass the Firecracker microVM or host-side network boundary. This proactive program aims to identify and patch potential agent-driven exploits in the sandbox infrastructure.
VercelAug 22Vercel lowered the cost of GPT-5.6 Sol on AI Gateway by applying its 50% discount to OpenAI's new, reduced list prices. Input costs dropped 20% and output costs fell 33% across all service tiers, including fast mode. These rates remain effective through September 18, 2026, and apply automatically to existing model identifiers without requiring code changes.
VercelAug 22Vercel added DeepSeek V4 Flash Vision Experimental to its AI Gateway. This experimental model accepts images alongside text with a 1M token context window and supports JPEG, PNG, GIF, and WebP formats. It integrates with coding agents like Cursor and Claude Code, reflecting provider pricing with no platform markup or inference fees.
v0Aug 22Vercel integrates Vercel Connect into v0, providing secure connections to over 100 third-party services including Slack, GitHub, and Salesforce. The system uses reusable team connectors and short-lived tokens to eliminate the need for long-lived provider secrets. This update enables v0-generated agents to interact directly with external data sources during task execution.
VercelAug 22Vercel launched Vercel Agent for Slack, bringing production-aware AI assistance into team conversations. The agent diagnoses incidents, reviews pull requests, and updates configurations directly within Slack threads. Changes require explicit human approval through a plan card before execution. This integration is available in public beta for Pro and Enterprise teams.
VercelAug 22Vercel Labs open-sourced fx, a tiny, native coding agent written in Zig. Optimized for research and embedding, the 7.8 MiB binary functions as a minimalist CLI that is model-agnostic, supporting Vercel AI Gateway, OpenAI Codex, and xAI Grok. It is designed for performance and embeddability in larger systems via WebAssembly and is currently available as an experimental release.
VercelAug 22Vercel added Z.ai's GLM 5.3 to AI Gateway, featuring enhanced vulnerability discovery and multi-step reasoning capabilities. The model serves as a specialized open-weight option for cybersecurity tasks, supporting function calling, structured output, and context caching. It provides a 1-million-token context window with no platform fees or markup on inference requests.
VercelAug 22Vercel integrated Cline into the AI SDK HarnessAgent interface, providing a unified runtime for coding agents. The new @ai-sdk/harness-cline adapter allows Cline to operate within the harness layer, supporting runtime switching without application code changes. It joins existing supported agents including Claude Code, Codex, and Grok Build.
VercelAug 14Vercel launched a CLI command that automatically configures nine popular coding agents, including Claude Code and Cursor, to route traffic through AI Gateway. This integration centralizes observability, spend budgets, and security controls like Zero Data Retention across 300+ models from 30+ providers, replacing the need for fragmented, agent-specific configurations.
VercelAug 13Vercel made Z.ai’s GLM 5.2 coding model free for all eve agents on AI Gateway through August 27, 2026. The model, featuring a 1-million-token context window, provides up to 500 tokens per second via Blackbox AI. This offer applies to the standard model variant and is now the default for new agents initialized via the CLI.
VercelAug 13Vercel added Google’s Gemini 3.7 Flash to its AI Gateway, offering the model at a 50% discount through December 31, 2026. The model improves performance in software engineering and agentic tasks, including more reliable tool-calling sequences. Developers can access the model via the Vercel AI SDK using the google/gemini-3.7-flash identifier.
VercelAug 13Vercel integrated Exa’s agentic search into AI Gateway and set it as the default web search tool for eve agents. The integration works with any AI Gateway model without requiring a separate Exa API key. Access to Exa through these services is free for all users through August 31, 2026.
VercelAug 13Vercel added the Grok Build harness adapter to the AI SDK. The @ai-sdk/harness-grok-build package integrates xAI’s coding agent into the unified HarnessAgent interface. This addition enables switching between coding agent runtimes without modifying application code.
VercelAug 13Vercel released the @ai-sdk/harness-acp package, a meta adapter that enables the AI SDK's HarnessAgent to connect with any Agent Client Protocol (ACP)-compatible harness. This update allows integration with agent runtimes that lack direct SDK adapters. For harnesses with existing direct adapters, Vercel recommends using those for tighter integration.
VercelAug 13Vercel introduced Vercel Managed Images, a set of versioned, open-source base images for Vercel Sandbox. The new default image runs on Ubuntu 26.04 and includes preinstalled coding agents like Codex and Claude, alongside Node.js and Python. These images replace previous Sandbox runtimes, providing automated nightly updates and support for custom, extended environments.
VercelAug 12Vercel deployed updated weights for DeepSeek V4 Pro on AI Gateway. Requests to the existing model identifier automatically receive the new version, while pinning to the specific identifier deepseek/deepseek-v4-pro-0813 ensures consistent access. The update maintains provider pricing with no platform markup or inference fees for all requests.
v0Aug 12Vercel integrates xAI’s Grok 4.6 into v0, the company’s AI-powered web application builder. The model is now available for selection within the v0 interface to assist in designing, iterating, and scaling full-stack web applications.
VercelAug 12Vercel launched an AI SDK software factory where specialized agents handle bug reproduction, feature implementation, and backporting. Humans retain control by reviewing and merging every change. Four weeks into production, the factory authors up to 35% of merged pull requests, closed 70% of issues in July, and reduced open bugs by approximately 25%.
VercelAug 12Vercel added xAI's Grok 4.6 to AI Gateway, featuring a 500K token context window and support for text and image inputs. The model includes configurable reasoning levels, defaulting to high. Existing Grok 4.5 traffic migrates to the new version via a CLI rewrite rule, eliminating the need for application redeploys.
VercelAug 12Vercel updated the Vercel CLI to support creating all 100+ Vercel Connect service connectors directly from the terminal. This update removes the requirement to switch to the dashboard for setup, as the configuration, attachment, and token requests for any supported service now occur entirely within the command line.
VercelAug 12Vercel’s August AI Gateway Production Index reports DeepSeek surpassed Google as the second-largest lab by token volume. Average token prices fell 13.6% as production workloads shifted toward lower-cost models. Anthropic maintained 65% of gateway spend despite higher pricing, while new open-weight models Kimi K3 and GLM 5.2 captured significant volume and revenue share for the first time.
VercelAug 9Vercel integrated its AI Gateway and Sandbox into the Hermes Agent. The AI Gateway provides inference access to over 200 models with spend observability, while the Sandbox runs agent commands in isolated microVMs. Configuration occurs through the Hermes setup wizard or CLI to manage model routing and secure execution environments.
VercelAug 9Vercel integrated ByteDance’s Seedance 2.5 video generation model into its AI Gateway. The model supports text-to-video generation and reference-based editing, accessible via the Vercel AI SDK using the model identifier bytedance/seedance-2.5. The integration enables programmatic generation of 30-second videos with precise reference control through the platform’s unified API.
Frequently asked questions