English static mirror for SEO/GEO · AI-assisted translation · Read Chinese original

11 Days, 64 Parallel Claude Instances: Rewriting 1 Million Lines of Bun from Zig to Rust

Forum topic · 小凯 · 2026-07-12

Summary

Developer Jarred Sumner announced on July 8, 2026 that Anthropic's Claude Fable 5 model completed a full rewrite of the Bun JavaScript runtime from Zig to Rust in 11 days. 64 Claude instances ran in parallel, generating over one million lines of code and costing roughly USD 165,000 in API fees. Sumner estimates a human team would have needed about one year for equivalent work. The motivation was Zig's persistent memory safety issues; Rust eliminates most of these at compile time. Bun v1.4.0 shipped in Canary, fixing 128 bugs and delivering a 2 to 5 percent runtime speedup. Anthropic acquired the Bun project in December 2025, so the API cost was internal rather than personal. The case is framed as Anthropic's first concrete demonstration of full-stack AI coding, combining a frontier model, the Claude Code IDE, and a JavaScript runtime as an integrated toolchain. Open questions remain about regression coverage, ecosystem migration costs, and whether the workflow transfers to projects without deep domain knowledge.

11 Days, 64 Parallel Claude Instances: Rewriting 1 Million Lines of Bun from Zig to Rust

> Event date: 2026-07-08 (blog post) / indexed by aihot on 2026-07-11 > Sources: IT之家 + aihot > Score: 71 > aihot link: https://aihot.virxact.com/items/cmrg3syhp00l7iha7a3ansg62 > Original link: https://www.ithome.com/0/975/469.htm

What Happened

On July 8, developer Jarred Sumner announced that with the help of Anthropic's Claude Fable 5 model, his team completed an unlikely engineering feat in 11 days at a cost of roughly USD 165,000: a full rewrite of the Bun JavaScript runtime from Zig to Rust.

Key figures:

  • 64 Claude instances ran in parallel for 11 days.
  • Over 1,000,000 lines of code were produced.
  • API cost: approximately USD 165,000 (about CNY 11.19 million).
  • Estimated human-team equivalent: ~1 year.
  • Primary motivation: Zig's memory errors and crashes were difficult to eliminate, whereas Rust catches most issues at compile time.
  • Bun v1.4.0 shipped as a Canary release, fixing 128 bugs and delivering a 2-5 percent runtime speedup.
  • Anthropic acquired the Bun project in December 2025, so the API bill did not land on Sumner personally.

    Why This Matters: A Full-Stack AI Coding Demonstration

    The event is best read as the first concrete realization of Anthropic's full-stack AI coding strategy. Unlike OpenAI's acquisition of Windsurf (an IDE asset), Anthropic bought a JavaScript runtime. Combined with Claude Fable 5 (model) and Claude Code (IDE), Anthropic now owns all three foundational layers of an AI coding workflow: model, IDE, and runtime.

    The 11-day, one-million-line volume is a deliberate mismatch with traditional engineering timelines. With 64 parallel Claude instances, work that would normally be sequential becomes parallel: each instance handles a module or feature, producing fine-grained artifacts such as "rewrite Bun's compile-time checks in Rust," "rewrite the HTTP server," and "rewrite the fs API."

    Cost comparison sharpens the point. USD 165,000 versus an estimated USD 2-5 million for a senior US engineering team translates to roughly a 12x cost reduction and a 50x or greater productivity multiplier. This is the first time a "scaled rewrite" has been quantified in dollars.

    Three Shifts in the Boundary of AI Coding

    1. From completing a function to refactoring million-line projects. The narrative has moved from inline copilots, to Cursor-style cross-file refactors, to language-level migrations. 2. From single-threaded assistants to AI workforce. 64 parallel instances turn Claude Code into a team rather than a helper. The differentiator is parallelism, not raw model intelligence. 3. From model pricing to full-stack strategy. The USD 165,000 is not Fable 5's price tag but Anthropic's "entry ticket" for a coding-tool experiment that can be replicated on any runtime (databases, web frameworks, compilers).

    Sumner's post does not disclose the ratio of AI-generated to human-guided lines, but a reasonable estimate is that 30-50 percent of the work was directly produced by Fable 5, with the remainder consisting of Sumner's review, prompting, and CI/CD validation. Even with that mix, the 11-day timeline is implausible without the model.

    Risks and Open Questions

  • Canary is not GA. Bun v1.4.0 is a Canary release; no regression pass rate, production compatibility data, or memory-leak measurements were published. The 2-5 percent speedup blends Rust's intrinsic benefits with net optimization from the rewrite in unclear proportions.
  • Ecosystem migration costs. Rewriting 1,000,000 lines is tractable; migrating the surrounding OSS dependencies, user code, and third-party plugins is not automated away. The timeline for v1.4.0 to fully replace v1.3's ecosystem position is undisclosed.
  • Transferability. The 11-day result may depend heavily on the Bun team's domain expertise (the authors know every module intimately). Whether Claude Fable 5 retains this efficiency on a foreign codebase is an open engineering question.
The most important caveat: this scaled rewrite is an internal Anthropic artifact. It demonstrates possibility, not a replicable SOP for outside teams attempting to rewrite legacy codebases with AI tools.

Tags

#ai-coding#claude-fable-5#bun-runtime#rust-rewrite#parallel-agents#anthropic#software-migration#agentic-engineering

This page is an English static mirror generated for search and AI citation. It may be a full translation or structured summary of the Chinese original. Canonical interactive discussion lives on the Chinese page: https://zhichai.net/topic/178379396