Vercel

Set spend caps on API keys for AI Gateway to control AI costs. Once a budget limit is exceeded, AI Gateway rejects further requests on that key until the budget resets or is raised. The cap applies to all AI Gateway providers and models running through the key.

Create budgeted API keys in the Vercel Dashboard via the AI Gateway API Keys page, or programmatically through the Vercel CLI:

vercel ai-gateway api-keys create --name <NAME> --budget <DOLLARS> --refresh-period <PERIOD>

Refresh periods can be daily, weekly, monthly, or none. Each period resets at the start of its window in UTC.

Claude Fable 5 from Anthropic is now available on AI Gateway. A Mythos-class model, Fable 5 is a notable step up over prior Claude models on long-running, ambiguous, multi-step tasks. The model sustains productive output across multi-day runs and dependably dispatches parallel sub-agents. Code review, bug-finding, and repository investigation are stronger, and first-shot correctness on complex problems is noticeably higher.

Fable 5 ships with blocking classifiers that refuse offensive cybersecurity, biology, and summarized-thinking extraction. Anthropic does not support Zero Data Retention; prompts and completions are retained for 30 days and are not used to train Claude.

To use Fable 5, set the model to anthropic/claude-fable-5 in the AI SDK.

Search domains using the Vercel CLI with the vercel domains search command. Supply a domain name to retrieve availability and price results for all supported TLDs.

Filter by TLD, apply sorting, and filter out unavailable domains:

vercel domains search acmesite --tld com --tld ai
vercel domains search acmesite --available
vercel domains search acmesite --order alphabetical
vercel domains search acmesite --format json
vercel domains search --help

Upgrade your Vercel CLI to version 54.10.1 to get started.

Vercel Sandbox now supports drives in private beta. Drives are persistent, attachable storage with a lifecycle independent from any sandbox.

Create a drive once, then mount it at a configurable path when starting a sandbox. When the sandbox stops, the drive remains available to attach to a later sandbox.

Install the beta SDK (@vercel/sandbox@beta) or beta CLI (sandbox@beta), then create and mount a drive.

Sandbox Drives are useful for:

• Keeping agent workspaces across disposable sandboxes
• Retaining cloned repositories, dependencies, and build outputs
• Managing data independently from the sandbox lifecycle

During the private beta, a drive can be mounted read-write by one sandbox at a time. Sandbox drives should not be used for production data while in private beta. Sign up to join the waitlist.

The skills.sh API is now available. Authenticate with your project's Vercel OIDC token and start querying more than 600,000 skills from across the open-source ecosystem.

Search for skills, pull detailed info on any one, check its security audit, and more. Vercel issues a short-lived token scoped to your team and project, rotated automatically, so there's no long-lived secret to leak or rotate. On each request, skills.sh verifies the token and applies a rate limit of 600 requests per minute per team and project.

Nemotron 3 Ultra from Nvidia is now available on Vercel AI Gateway. This is an open Mixture-of-Experts reasoning model built for orchestrating long-running agent workflows, with a 1M token context window.

Key features:

• Targets multi-turn agent workflows: planning, tool use, sub-agent delegation, and error recovery
• Throughput up to 350 tokens per second
• Up to 30% lower cost on agentic tasks
• Use model nvidia/nemotron-3-ultra-550b-a55b in AI SDK

AI Gateway provides unified API for calling models, tracking usage and cost, configuring retries, failover, and performance optimizations. Includes custom reporting, Zero Data Retention support, and dynamic provider sorting.

Vercel has updated its Terms of Service and Marketplace terms to clarify shared responsibility when actions on your account may be taken by AI, whether Vercel's own or third-party tools. Key updates include:

AI Functionality and Third-Party Tools: New concepts introduced to clarify responsibilities around AI-powered services (v0, Vercel Agent, WAF, AI Gateway) and non-Vercel tools granted account access. You control how AI Functionality behaves, are responsible for reviewing its outputs, and responsible for any Third-Party Tool actions on your behalf.

Billing Flexibility: Fees are no longer exclusively tied to subscription plans. Services like AI Gateway credits can be purchased independently. Charges may occur at purchase or top-up. High or suspicious usage may trigger early charges ahead of regular billing cycles.

Enterprise Managed Accounts: Enterprise customers can now claim email addresses on verified domains associated with existing Vercel accounts.

Representations: Two new customer confirmations—rights to provide third-party credentials and confirmation of non-OFAC sanctions status.

Marketplace: Authorized Users concept introduced, binding actions to the customer.

Privacy Policy: Data Privacy Framework updated with VeraSafe as private complaint handler.

Grok Imagine Video 1.5 from xAI is now available on AI Gateway. The model generates video from an input image with synchronized audio in a single pass.

This release improves audio quality, prompt following, and photorealism. Face accuracy and character consistency are stronger across longer sequences, with better lighting and physical realism. Reference image support has been expanded to give more control over visual style and subject.

Use the model by setting it to xai/grok-imagine-video-1.5-preview in the AI SDK. You can chain an image model with Grok Imagine Video 1.5 to generate a still and animate it in one flow. Try this model directly in the AI Gateway Playground.

AI Gateway provides a unified API for calling models, tracking usage and cost, and configuring retries, failover, and performance optimizations. It includes built-in custom reporting, Zero Data Retention support, dynamic provider sorting by latency and cost, and more.

Vercel Blob now supports time-bound signed URLs for scoped, secure access without exposing your entire Blob store. Each signed URL is scoped to a single operation (put, get, head, or delete), a single pathname, and an expiry up to 7 days. Features include:

• Direct uploads from the browser: Upload URLs (put) support multipart for streaming large files straight to Blob storage
• Conditional deletes: Delete URLs accept an ifMatch option so deletion only applies if the object hasn't been overwritten since the URL was signed
• OIDC integration: Your server authenticates via OIDC, generates signed tokens, and produces narrowly scoped URLs for the browser, keeping your BLOB_READ_WRITE_TOKEN server-side

Update @vercel/blob to 2.4.0 or later.

Monorepos that deploy many projects can now configure all of their project's Git settings more conveniently. Previously, you had to click through to every project's settings and apply the same setting consistently. Now you can configure commit status, repository_dispatch events, and other settings for all projects in one place via project settings.

Qwen 3.7 Plus from Alibaba is now available on Vercel AI Gateway. Both Qwen 3.7 Plus and 3.7 Max are free for paid AI Gateway users through 6/4/26 12:00pm PT.

The model unifies vision and language into a single agent foundation with capabilities spanning GUI and CLI operation, coding and productivity workflows with full-modality input, and visual agent tasks including perception and reasoning. Use alibaba/qwen-3.7-plus as the model in the AI SDK.

Elastic build machines now monitor your build's memory usage and automatically adjust to prevent out-of-memory (OOM) failures:

• If your build is fast but memory-intensive, Vercel will no longer downgrade you to a smaller machine
• If your build is close to running out of memory, it will automatically upgrade to a higher tier
• If your build fails due to an OOM error, the next deployment will automatically run on a higher tier

Thresholds are set conservatively to balance deployment reliability and cost. Enable elastic builds in your team or project settings.

MiniMax M3 is now available on Vercel AI Gateway. M3 is MiniMax's first model with a 1M-token context window and native multimodality, built around MiniMax Sparse Attention (MSA). M3 improves on software engineering, terminal-based tool use, and agentic web browsing, and is tuned for multi-turn collaboration.

Use in the AI SDK by setting model to minimax/minimax-m3. Supports multimodal input with text and images.

AI Gateway provides a unified API for calling models, tracking usage and cost, and configuring retries, failover, and performance optimizations. Includes custom reporting, Zero Data Retention support, dynamic provider sorting by latency & cost, and more.

Vercel Sandbox now supports installing and running Docker inside a sandbox. An agent can build containers, install system packages, and modify files without touching your host system. Install Docker, start the daemon, and serve containerized applications directly within the sandbox environment.

Function invocations are moving from package-based to per-unit pricing for Pro and new Enterprise customers. Starting with your next billing cycle, you'll be billed per unit at $0.0000006 per invocation (previously $0.60 per 1M invocations) for Pro customers. Per-unit billing scales more smoothly across team sizes and usage patterns and helps teams on Pro use function invocations without immediately consuming a large portion of the included monthly usage credit.

Starting April 29th, the maximum retention policy for Hobby plans will be capped at 30 days. Deployments outside your retention window will be automatically removed. This excludes your 10 most recent production deployments and any aliased deployments, which continue to be preserved regardless of retention settings.

Pro and Enterprise plans are not affected.

Learn more about Deployment Retention.

GPT-5.5 is now available on Vercel AI Gateway. Two variants: GPT-5.5 and GPT-5.5 Pro, both tuned for long-running agentic work across coding, computer use, knowledge work, and scientific research.

GPT-5.5 excels at agentic coding and long-horizon work with strong context retention. GPT-5.5 Pro targets demanding, multi-step work prioritizing response quality over latency, with gains in business, legal, education, data science, and technical research workflows.

Use model openai/gpt-5.5 or openai/gpt-5.5-pro in the AI SDK. AI Gateway provides unified API access with custom reporting, observability, Bring Your Own Key support, and intelligent provider routing with automatic retries.

DeepSeek V4 is now available on Vercel AI Gateway with 1M token context window default across both variants: DeepSeek V4 Pro and DeepSeek V4 Flash.

DeepSeek V4 Pro focuses on agentic coding, formal mathematical reasoning, and long-horizon workflows, handling feature development, bug fixing, and refactoring across stacks with tool use for MCP workflows and agent frameworks. DeepSeek V4 Flash performs close to V4 Pro on reasoning with smaller parameters for faster responses and lower API cost, ideal for high-volume and latency-sensitive workloads.

Use model deepseek/deepseek-v4-pro or deepseek/deepseek-v4-flash in the AI SDK.

GPT Image 2 is now available on Vercel AI Gateway. OpenAI's newest image model supports detailed instruction following, accurate object placement and relationships, and dense text rendering across multiple aspect ratios up to 2K resolution.

The model renders fine-grained elements including small text, iconography, UI elements, dense compositions, and subtle stylistic constraints. Non-English text is supported and reads coherently. GPT Image 2 produces photos, cinematic stills, pixel art, manga, and other visual styles with consistency in texture, lighting, composition, and detail—suited for game prototyping, storyboarding, marketing creative, and asset generation.

Use model openai/gpt-image-2 in the AI SDK or try the model playground.

Kimi K2.6 from Moonshot AI is now available on Vercel AI Gateway. The model focuses on long-horizon coding tasks with generalization across languages such as Rust, Go, and Python, and across front-end, devops, and performance optimization work. K2.6 can turn simple prompts into complete front-end interfaces with structured layouts.

For autonomous, proactive agents running continuously across multiple applications, K2.6 improves on API interpretation, long-running stability, and safety awareness during extended research tasks.

Use model moonshotai/kimi-k2.6 in the AI SDK.