Bun 1.4 rewrite from Zig to Rust and runtime performance

Jarred Sumner converted 960,000 lines of Zig into Rust in six days. He used Claude to perform the file-by-file translation. This rewrite defined Bun 1.4, which released on August 20, 2026, after testing inside Anthropic’s Claude Code. Anthropic acquired Bun in December 2025 to support its AI coding infrastructure. The rewrite aims to improve stability at the boundaries where the runtime interacts with C++ libraries like JavaScriptCore. I find the speed of this AI-driven transition impressive. However, the lack of a human-led architecture redesign creates a new kind of risk. The rewrite of 960,000 lines of Zig into Rust took only six days via Claude, a feat that occurred after Anthropic acquired the project in December 2025 to ensure its AI agents had high-performance infrastructure. Sumner originally chose Zig because it provided C-level performance with easier learning than Rust. He valued Zig’s "comptime" feature for evaluating arbitrary code at compile time. He also used "defer" to handle resource cleanup at the end of a scope.

Performance and the runtime competition

Bun 1.4 provides measurable efficiency gains in production environments. The runtime starts 50% faster on Linux and reduces idle CPU usage by five times. Memory usage fell between 13 and 48 percent depending on the framework. I compared Bun to Node.js and Deno across 14 microbenchmark tests on an Apple M4 machine.

Bun 1.4 reached 95 percent Node.js compatibility by adding 1,517 tests from the Node.js suite. Deno 2 provides better security because it sandboxes every operation. Node.js maintains the highest level of compatibility for legacy enterprise systems. Bun provides 10x faster package installs than Node.js and 35% faster cold starts in AWS Lambda. I would skip the hype about Bun killing Node.js, because Node.js remains competitive in bulk hashing and long-running stability. Deno remains the top choice for security-critical greenfield projects using TypeScript. Bun also excels at server-style throughput, especially in HTTP and large JSON parsing. Deno led in the async scheduling tests and the tight arithmetic loop.

Is the runtime rewrite stable?

The shift to Rust does not stop all memory issues. I saw MongoDB connections in Bun leak memory over time. Next.js applications on Bun also show steady memory growth and higher CPU consumption. Many developers treat Bun as a short-lived process for serverless functions to avoid these issues. You should move the package manager and the test runner before you move the HTTP server.

The release of version 1.4.2 on September 5, 2026, fixed a regression in AsyncLocalStorage. This Node API carries values across asynchronous boundaries and supports request-scoped concerns like logging and authentication. The regression shows that rewriting a runtime can create quiet failures in the async plumbing. Does the speed of an AI-driven rewrite justify the risk of silent context loss? Developers who encounter these memory leaks often treat Bun like PHP. They use it only for short-lived processes.