Bun rewrote its runtime from Zig to Rust in 11 days using 64 Claude instances

Bun disclosed on July 8, 2026 that it rewrote the core of its JavaScript runtime from Zig to Rust in 11 days, with a single engineer coordinating roughly 64 parallel Claude instances. Those instances did most of the work.

The rewrite ran from May 3 to May 14, 2026, using Anthropic’s Claude Fable 5 model inside Claude Code, arranged as an implementer-and-reviewer workflow across the parallel instances, Bun said in a blog post. The disclosure lands two months after the work finished and is the first public account of it.

The scale is the story. Bun said the effort produced 6,502 non-merge commits and a net diff of just over 1,009,272 added lines, and cost about $165,000 at standard API pricing. The token accounting was large: 5.9 billion uncached input tokens, 690 million output tokens and 72 billion cached input-token reads.

Bun said the rewrite passed a test suite of 1,386,826 assertions across 60,624 tests with none skipped or deleted, and that the Rust build cut peak memory use from 6.7GB to 609MB versus the prior Zig version.

These are Bun’s own figures for one project on a mature, well-tested codebase, not a controlled benchmark, so they do not show that agentic rewrites generalize to less-covered code. But few teams have published cost and token totals at this scale, which makes the numbers a rare concrete data point on what large agentic refactors actually consume.

Whether the Rust runtime reaches a stable release will tell readers if the rewrite holds up outside the test suite.