Bun Rewritten from Zig to Rust in 11 Days Using Claude Fable 5 with 64 Parallel Instances
> Event date: 2026-07-08 (blog post published) / indexed by aihot 2026-07-11 > Source: IT之家 + aihot > aihot link: https://aihot.virxact.com/items/cmrg3syhp00l7iha7a3ansg62 > Original article: https://www.ithome.com/0/975/469.htm
What Happened
Developer Jarred Sumner announced on July 8 that, with the help of Anthropic's Claude Fable 5 model, he completed an "unlikely" engineering feat in 11 days at a cost of $165,000: rewriting the JavaScript runtime Bun entirely from Zig to Rust.
Key figures:
- 64 parallel Claude instances running for 11 days;
- More than 1 million lines of code written;
- API costs of roughly $165,000;
- A human team of the same scale would need about 1 year, per Sumner's estimate;
- Main motivation: Zig's memory errors and crashes were hard to fully fix, while Rust catches most issues at compile time;
- Bun v1.4.0 has shipped as a Canary release with 128 bug fixes and 2–5% speed improvements.
- Canary ≠ GA. No regression-test pass rates, production compatibility, or memory-leak data were published. How much of the 2–5% speedup comes from Rust itself vs. rewrite-level optimization is unclear.
- Ecosystem reverse-cost. The rewrite is done, but ecosystem dependencies, user migrations, and third-party plugin adaptations are not simplified by AI. No timeline for fully replacing v1.3's ecosystem position was given.
- Transferability. The success may depend heavily on the Bun team's domain expertise as the project's original authors. Whether Fable 5 retains this efficiency on an unfamiliar project is an open engineering question.
Anthropic acquired the Bun project in December 2025, so the $165,000 API cost does not fall on Sumner personally.
Analysis
The author's view: this is the first concrete demonstration of Anthropic's "full-stack AI coding strategy."
Anthropic's December 2025 acquisition of Bun differs from OpenAI's Windsurf deal — OpenAI absorbed a coding IDE asset, while Anthropic absorbed a coding runtime asset. Claude Code is an IDE; Bun is a JavaScript runtime — both are foundational tools in the AI coding workflow. Anthropic now builds the model (Fable 5), the IDE (Claude Code), and the runtime (Bun), so Fable 5 was validated on the Bun rewrite before release while Bun becomes a natural foundation for Claude Code.
The mismatch in scale is the key: 11 days would barely suffice for requirements analysis under traditional engineering, but 64 parallel Claude instances turned a linear process into a parallel one, each instance handling a module or feature — rewriting Bun's compile-time checks, HTTP server, fs API, and so on.
The $165,000 API cost looks alarming but is actually cheap: a human team would cost at least $2–5 million per year (senior US engineers), a 12x cost difference and arguably 50x+ productivity gain. This is the first time large-scale AI-assisted rewriting has been quantified in real money.
Why It Matters
Three layers redefine AI coding's boundaries:
1. From "function completion" to "million-line project refactoring." Early AI coding was Copilot writing the next line; recently Cursor did cross-file refactoring; Bun crosses into million-line language migration — a compound result of model capability plus harness engineering. 2. From "single-threaded" to "AI workforce." 64 parallel instances mean Claude Code can be used as a *team*, not a personal assistant. The next dividing line is parallelism, not just stronger models. 3. From "model cost" to "full-stack strategy." The $165,000 is an entry ticket to an AI coding tooling experiment post-acquisition; the paradigm could be replicated to any runtime (databases, web frameworks, compilers).
Notably, Sumner's post does not give the AI-generated vs. human-guided ratio within the 1 million lines — plausibly only 30–50% was fully AI-generated, with the rest involving review, prompt guidance, and CI/CD validation. Even so, 11 days remains an impossible-sounding task; the scarce resource is the person who can direct the AI, not the AI itself.