Vercel has recently launched a significant set of new features, many of which address long-standing developer pain points, particularly in the realm of AI-driven applications and performance.
Overview of Vercel's New Features 02:18
- Vercel has shipped a multitude of new features, including a new pricing model, queues, sandboxing for code, and improved captcha handling.
- These updates are seen as a significant leap forward, as previous Next.js conferences were relatively light on new features for both Next.js and Vercel.
Active CPU Billing 03:09
- Previously, serverless platforms billed based on "gigabyte-hours," meaning an instance was alive for the duration of a request, which was efficient for short requests (100-200ms).
- This model became problematic for AI generation requests, which are high duration (seconds to minutes) but have extremely low CPU usage (nanoseconds).
- Vercel's new "Active CPU Pricing" model addresses this by charging only when the code is actively using CPU, similar to Cloudflare's "net CPU" billing.
- This change is expected to significantly reduce costs for AI workloads, potentially making compute bills 400 times lower than inference bills for the speaker's projects.
- Vercel now charges $0.128 per hour for active CPUs, $0.01 per gigabyte-hour for provisioned memory, and $0.60 per million requests.
- While Cloudflare may still be slightly cheaper per hour for raw CPU, Vercel offers benefits like faster CPU, Node compatibility, and superior developer experience, making the overall cost gap (roughly 2x more expensive for Vercel) acceptable.
- Vercel has also formally moved middleware and edge workers to run on its core compute engine.
Sandboxing for Code 31:59
- Vercel Sandbox allows developers to securely run untrusted code, such as AI-generated or user-submitted code, in isolated and ephemeral environments.
- It's available as a standalone SDK that can be used even on non-Vercel platforms like Cloudflare or AWS.
- This feature is crucial for safely executing code that AI can write but needs to interact with the real world.
- Vercel's sandbox provides Node.js compatibility, unlike Cloudflare Workers which have strict limitations on binary size and memory, making complex Node.js APIs challenging.
Queues 38:18
- Vercel has introduced a message queue service, currently in limited beta, designed to offload long-running tasks to the background.
- This allows applications to handle slow operations (e.g., AI video processing, image generation) without making users wait, and improves reliability with built-in retries and failure handling.
- Key features include PubSub messaging, streaming support for large payloads, streamlined OAuth, and a fully type-safe TypeScript SDK.
Bot ID (Captcha Killer) 43:09
- Vercel's "Bot ID" is an invisible bot filtering solution for critical routes, designed to combat sophisticated bots that mimic human behavior.
- It operates silently by injecting obfuscated code and collecting thousands of signals, without relying on traditional, easily faked methods or intrusive user challenges.
- Pricing offers a "Basic" mode, free for all plans, and a "Deep Analysis" mode for Pro and Enterprise plans at $1 per thousand requests, competitively priced with existing captcha solutions like HCaptcha and Recaptcha.
- This solution aims to provide a much better developer and user experience compared to traditional captchas, which often lack modern DX and can have high false positive rates.
AI Gateway 56:29
- The AI Gateway provides a single endpoint to access various AI models (OpenAI, XAI, Anthropic, Google) with improved uptime, faster responses, and reduced vendor lock-in.
- It addresses common issues with direct model provider APIs, particularly Anthropic, which suffers from poor reliability, restrictive rate limits (e.g., 2 requests per minute for high-token inputs on Tier 4), and unfavorable cost negotiations (e.g., 5% discount for $240,000/year committed spend).
- Similar to Open Router, Vercel's AI Gateway can route traffic to the most reliable and fastest provider, abstracting away underlying infrastructure and rate limits.
Other Notable Updates 66:19
- Rolling Releases: Allows developers to deploy new application versions to a percentage of users, facilitating phased rollouts and large overhauls.
- Microfrontends: Enables breaking down large applications into smaller, independently deployable units, allowing different teams to build, test, and deploy their parts autonomously.
- Vercel Agents: An AI assistant integrated into the Vercel dashboard designed to analyze app performance and security data, identifying anomalies, causes, and recommending specific actions.