Vercel

Vercel launched Fluid Compute, a serverless runtime architecture that dynamically scales CPU and memory with zero cold starts and active connection pooling.

Klaviyo utilized Vercel for Enterprise Apps and Agents to enable 512 employees to deploy 356 internal applications within two weeks. The platform leverages Secure Compute and Vercel Passport to enforce SSO-gated, private-by-default deployments for non-technical staff.

Vercel has transitioned its security vulnerability reporting from a private, invite-only model to a unified public bug bounty program hosted on HackerOne. This consolidation covers all Vercel platform products and open-source projects under a single submission portal.

Vercel has introduced memory usage tracking for Sandboxes, providing visibility into average, P75, and P95 memory consumption. This data is accessible via the Vercel dashboard charts and the Vercel CLI using the vercel.sandbox.memory_used_bytes metric.

Vercel has integrated the Pixel Canary model into its AI Gateway, currently available in a free stealth phase. The model achieves a 96.8% success rate on Next.js evaluation tasks when provided with AGENTS.md documentation.

Vercel's skills.sh registry has reached one million agent skills within seven months of its launch. The registry currently facilitates nearly 280 million total installs for AI agent instructions.

Vercel has introduced the vercel/vcr-action/login GitHub Action to enable image pushes to Vercel Container Registry (VCR) using OIDC tokens. This eliminates the need for long-lived registry credentials by exchanging workflow OIDC tokens for short-lived Vercel access tokens.

Vercel AI Gateway now supports Google's Gemini 3.8 Flash-Lite TTS and Gemini 3.8 Flash TTS models for text-to-speech generation. These models enable multi-language narration with specific controls for tone, pacing, and character design.

Vercel has eliminated the hard caps on the number of Blob stores allowed per account, moving from fixed limits of 100 to 1,000 stores to an unlimited model. Store creation is now billed as a Blob Advanced Operation, costing $5.00 per million operations on the Pro plan.

Vercel Connect now enables TanStack AI agents to call OAuth-protected MCP servers without manual credential management. Developers can utilize the new @vercel/connect/tanstack-ai subpath to implement secure, automated authentication for MCP requests.

Vercel AI Gateway now supports the Xiaomi MiMo V2.6 model family, including Pro, Flash, and Pro UltraSpeed variants. These models feature a 1M token context window and support for native multimodal inputs including text, image, audio, and video.

Vercel has updated its dashboard, REST API, and CLI to display specific billable duration and CPU minute metrics for deployments. Users can access these granular usage statistics by updating to Vercel CLI version 59.23.1 or later.

Vercel Connect now includes a managed connector for Microsoft Teams. Creating one gives your organization a Teams bot that your apps and age...

Vercel AI Gateway now supports Anthropic Claude Opus 5.5, which offers 30% faster performance and 40% lower costs compared to Opus 5. The update introduces breaking API changes including mandatory adaptive thinking and the deprecation of forced tool use.

Vercel has introduced persistent storage Drives for Sandbox environments, allowing data to persist across different sandbox instances. The feature supports up to four mounts per sandbox with storage capacities configurable up to 16 TiB.

Vercel AI Gateway now supports the integration of OpenAI's GPT-6 Sol and GPT-6 Luna models. These models are accessible via the AI SDK and OpenAI-compatible APIs for professional coding and agentic workflows.

Vercel and Next.js disabled AVIF image optimization across their platforms to mitigate a remote code execution (RCE) vulnerability found in the upstream libheif library. This mitigation was implemented following the discovery of the vulnerability in August 2026, pending the release of patched upstream dependencies.

Vercel has integrated the Jev probabilistic decision model from TypeSafe AI into its AI Gateway. The model is designed to return structured, typed outputs for software decision-making workflows.

v0 now enables the installation of private packages by reading NPM_TOKEN or NPM_RC credentials from Vercel shared environment variables. This integration allows teams to securely access internal component libraries and private registries directly within the v0 development environment.

The mcp-handler library has been updated to version 2.2.0 to include experimental support for the WebMCP standard. This allows developers to expose existing MCP tools to in-browser agents via a single script tag.

Vercel has expanded its Spend Management feature to Enterprise Flexible Commitment plans at no additional cost. Users can now set budgets to trigger email notifications, webhooks, or automatic production deployment pauses based on metered usage.

OpenAI released the GPT-6 Astra model on September 3, 2026, capturing 7.7% of total gateway spend in its first twelve days. The model launched at the same price point as Anthropic's Fable 5.1 but secured double the market share.

The skills CLI version 1.7.0 introduces support for Notion-hosted agent skills, allowing users to install skills directly from Notion pages. This integration eliminates the requirement for Git repositories by utilizing the new Notion Agent Skills API.

Vercel Sandbox now supports running Harbor evaluations, including Terminal-Bench, by utilizing isolated Firecracker microVMs. This integration requires Harbor version 0.22.0 or later and enables parallelized benchmarking across multiple AI models via AI Gateway.

Vercel now allows developers to enable Turbo build machines on a per-deployment basis using Git commit messages, Vercel CLI, or the REST API. This provides temporary resource scaling for specific builds without requiring global project setting modifications.

Vercel CLI version 59.16.0 enables sub-second deployments for static artifacts by bypassing the build step. This feature supports directories containing up to 10 HTML or Markdown files with a maximum total size of 5 MB.

Vercel AI Gateway now supports the GLM 5.3 FlashX model, enabling high-speed inference at approximately 200 tokens per second. This integration allows developers to utilize the model within coding agents and interactive applications via the unified AI Gateway API.

Vercel now promotes FastAPI frontends and static files to the CDN at build time, bypassing Vercel Functions for these requests. Users can control this behavior via the tool.vercel.fastapi.static configuration in pyproject.toml.

The Vercel AI SDK harness layer now supports GitHub Copilot integration through the new @ai-sdk/harness-github-copilot adapter. This allows developers to swap coding agents using the HarnessAgent interface without modifying application code.

Vercel has introduced a new permission layer for Vercel Connect, allowing Pro and Enterprise teams to restrict connector management to specific roles. This feature enables Owners to limit the creation and modification of connectors to users with the Owner role or the new Connector Manager extended permission.

Tailscale transitioned its internal AI routing layer from direct individual provider API integrations to Vercel AI Gateway. This migration enables unified cost tracking, usage monitoring, and automated zero data retention (ZDR) compliance across all model requests.

Featured migrated 374 Sanity sites from AWS Elastic Beanstalk to Vercel and adopted the Vercel Workflow SDK for long-running backend jobs. The team also integrated Vercel AI SDK and AI Gateway to manage traffic across 17 different AI models.

Vercel has doubled the storage capacity for all Sandbox environments from 32 GB to 64 GB. This update applies to both Vercel Managed Images and custom images, including those using the deprecated runtime property.

Vercel has integrated the inclusionAI Ling 3.0 Flash Sante model into its AI Gateway platform. The model is available for free usage until October 4, 2024, via two specific model IDs.

Vercel has integrated the OpenAI GPT-6 Astra model into its AI Gateway infrastructure. Developers can now access this model via the 'openai/gpt-6-astra' identifier for use in coding agents and automated workflows.

Cursor Cloud Agents can now utilize Vercel Sandbox as an execution environment via the Cursor Self-Hosted Machines API. This integration leverages Vercel Functions and Vercel Workflow to manage agent lifecycles within isolated Firecracker microVMs.

Vercel has introduced a 'Basic' build machine tier featuring 2 vCPUs and 8 GB of memory for Pro and Enterprise projects. This tier is priced at $0.007 per build minute, matching the existing cost structure for Hobby accounts.

Vercel has expanded its free domain offer for Pro teams to include the .app and .dev top-level domains (TLDs). This update increases the total number of eligible TLDs available for a one-year free registration from six to eight.

Meta's Muse Spark 1.3 model is now available on Vercel AI Gateway with a 1M token context window and support for text, image, and PDF inputs. The release includes both standard and contributor pricing tiers, with the latter utilizing user data for model training in exchange for lower costs.

Vercel unveiled Fluid Compute architecture, replacing traditional static worker slots with dynamic memory bursting and sub-10ms warm invocations.