Dynamic Workers Open Beta Brings Faster AI Code Execution

Cloudflare has opened Dynamic Workers to all paid Workers users, positioning it as a lightweight alternative to traditional containers for executing AI-generated code. The feature uses V8 JavaScript engine isolates instead of conventional Linux containers, and Cloudflare says this design delivers startup speeds about 100 times faster while using memory 10 to 100 times more efficiently.

Cloudflare described Dynamic Workers as a way to run AI-generated code inside secure, lightweight isolates. The company also said this approach is 100 times faster than traditional containers.

How Dynamic Workers Work in the Workers Runtime

Parent Workers Can Launch Additional Workers at Runtime

Dynamic Workers let a parent Worker create additional Workers during runtime and load arbitrary code on demand inside sandboxed environments. Instead of starting a full container or virtual machine for every execution, the platform creates V8 isolates. These are lightweight execution contexts that share the underlying JavaScript runtime and can start in milliseconds.

Two Dynamic Worker Loading Modes

The feature includes two ways to load code:

  • A one-time execution call for ephemeral tasks
  • A cached mode that keeps a Dynamic Worker warm for repeated requests

This gives developers a way to choose between short-lived execution and a warmer runtime for ongoing usage.

Companion Libraries for Dynamic Workers

Cloudflare released three related libraries alongside the open beta.

@cloudflare/codemode

This library allows LLMs to write and execute TypeScript that orchestrates multiple API calls in a single pass.

@cloudflare/worker-bundler

This package handles resolving npm dependencies at runtime.

@cloudflare/shell

This library provides a virtual filesystem with persistent storage.

Code Mode and AI Agent Orchestration

One of the main use cases Cloudflare is highlighting is "Code Mode." In this setup, AI agents write code to orchestrate tool calls instead of moving through those calls one step at a time. According to Cloudflare, this method can reduce inference tokens and cost by as much as 80 percent.

Dynamic Workers Pricing During Open Beta

Dynamic Workers are priced at $0.002 per unique Worker loaded per day, in addition to standard Workers pricing for requests and CPU time. Cloudflare said the per-Worker daily charge is not active yet during the beta period.

Market Position in AI Agent Sandboxing Infrastructure

The launch places Cloudflare in a growing market for AI agent sandboxing infrastructure. Companies including E2B, Modal, Docker, and Vercel are also competing to provide secure execution environments for autonomous AI systems.

Cloudflare’s strategy centers on the isolate model it has refined over nearly a decade. That model was originally built to run thousands of tenants per machine across its global edge network. Cloudflare’s position is that this gives it a structural advantage in both cost and speed compared with container-based and microVM-based alternatives.

Dynamic Worker Loader Background

Kenton Varda, a key architect of the Workers platform, previewed the underlying Dynamic Worker Loader technology in September 2025. He described it as enabling Workers to spawn new isolates from code specified dynamically at runtime, with capability-based sandboxing features.

FAQ

What are Cloudflare Dynamic Workers?

Dynamic Workers are a feature for paid Workers users that lets a parent Worker launch additional Workers at runtime to execute AI-generated code inside secure, lightweight isolates.

How are Dynamic Workers different from traditional containers?

They use V8 JavaScript engine isolates rather than conventional Linux containers, which Cloudflare says makes startup about 100 times faster and memory usage 10 to 100 times more efficient.

How much do Dynamic Workers cost?

The pricing is $0.002 per unique Worker loaded per day, plus standard Workers charges for requests and CPU time. Cloudflare said the per-Worker daily fee is not active during the beta period.