Anthropic Quietly Shipped Bun v1.4.0 in Rust to Millions of Claude Code Users
Claude Code v2.1.181 runs a pre-release Rust port of Bun, cutting startup time 10% on Linux , a runtime bet extension developers need to track.
1. Anthropic Quietly Shipped Bun v1.4.0 in Rust to Millions of Claude Code Users
Claude Code v2.1.181, released June 17th, ships with a Rust-rewritten version of Bun that has not yet appeared in any public tagged release. The most recent public Bun release on GitHub is v1.3.14, dated May 12th. Claude Code bundles v1.4.0. Simon Willison confirmed this on July 19th by running strings against the Claude binary and finding 563 .rs source filenames embedded in the binary alongside the version string. Jarred Sumner, Bun's creator, noted the Rust port delivered a 10% startup improvement on Linux. Otherwise, the rollout was quiet enough that "barely anyone noticed."
This is a meaningful infrastructure decision, not a footnote. Anthropic is now running pre-release, canary-tier runtime code across what Sumner describes as millions of devices. That is an unusual distribution channel for unreleased software: Claude Code as a de facto Bun canary fleet. For GitHub Copilot CLI and Cursor, which compete directly with Claude Code on developer workflow integration, Anthropic is compounding its distribution advantage with a faster cold-start. A 10% startup improvement is modest in isolation, but startup latency is exactly what developers feel when invoking an AI coding tool dozens of times per session. Extension developers and anyone benchmarking Claude Code against alternatives need to account for the fact that Claude Code's runtime baseline is now ahead of what Bun's public releases offer.
The broader pattern is worth tracking. Anthropic shipping a pre-release Bun build suggests a closer collaboration with Bun's team than a typical dependency relationship. The Rust port has since landed in Bun's canary channel via bun upgrade --canary, meaning Claude Code effectively beta-tested the Rust rewrite at scale before any public announcement. Watch whether this tightens into a formal partnership, and whether other AI coding tools start specifying runtime versions as a performance differentiator rather than leaving it implicit.