Detailed Analysis
Anthropic's Claude Code version 2.1.113 introduced a significant architectural shift by transitioning from a JavaScript-based npm bundle to a native executable, a change that carries meaningful consequences for users running older hardware. This migration was precipitated by an npm deprecation announcement in January 2026, which effectively compelled Anthropic to move toward a compiled binary distribution model. The immediate casualty of that transition is compatibility with pre-AVX2 processors — a class of CPUs that predates Intel's Haswell microarchitecture (circa 2013) and similarly aged AMD chips — meaning that a segment of older Mac users can no longer install or run Claude Code version 2.1.113 or later. Anthropic has acknowledged the situation directly in the GitHub issue thread, confirming that there are no current plans to restore support for pre-AVX2 platforms. As a stopgap, affected users have been directed to pin their installations to version 2.1.112, the last release distributed as a JavaScript bundle, available via the npm registry at `@anthropic-ai/claude-code`.
The timing of the compatibility break is particularly unfortunate for affected users. Anthropic published an engineering postmortem on April 23rd — linked within the same GitHub thread — detailing bug fixes that were incorporated into recent Claude Code releases. Those fixes are now inaccessible to users on pre-AVX2 hardware, who are effectively frozen at version 2.1.112 and cannot benefit from stability or correctness improvements made in subsequent versions. This creates a two-tiered user experience in which some users receive a progressively better product while others are stranded on an older, buggier baseline with no upgrade path in sight.
The shift from a JavaScript runtime to a native executable reflects a broader industry trend in developer tooling, where performance and distribution simplicity increasingly favor compiled binaries over interpreted or JIT-compiled code. Native executables can take advantage of platform-specific CPU instruction sets like AVX2 — which enable vectorized, parallel computation — making them faster and more efficient for computationally intensive tasks such as those involved in AI-assisted coding workflows. However, this optimization comes at a direct cost to backward compatibility. By compiling against AVX2 as a baseline, Anthropic implicitly sets a hardware floor that excludes machines that, while aging, may still be in active professional use. The research context notes that similar AVX2 baseline requirements have appeared in other AI inference tools such as vLLM, suggesting this is an emerging industry norm rather than an Anthropic-specific idiosyncrasy.
The broader significance of this development lies in what it signals about the evolving deployment strategy for AI-native developer tools. Claude Code, which began as a cloud-API-backed tool requiring minimal local compute, is now shipping in forms that impose local hardware constraints — a marker of increasing local execution demands or distribution infrastructure changes. As AI coding assistants mature and take on more sophisticated agentic behaviors, the pressure to leverage native performance capabilities will likely intensify, making hardware floor decisions like the AVX2 baseline a recurring point of friction for users on older machines. For Anthropic, the challenge going forward will be balancing the engineering efficiencies of native distribution against the accessibility concerns of a user base that spans a wide range of hardware generations.
Read original article →