The install size of the latest version of Claude Code is now 45% smaller.
According to Jarred Sumner, a member of Anthropic's technical staff, optimisations in version 2.1.248 of Claude Code shrank the package from 376 MB to 207 MB on Apple Silicon Macs and from 392 MB to 224 MB on Linux x64.
The biggest change is in the JavaScript bytecode bundled into Claude Code. “To make Claude Code start faster, source code is parsed ahead of time into bytecode,” Sumner posted on X.
In earlier versions, the resulting bytecode was around nine times larger than the JavaScript source code it was generated from. “After several optimisations to JavaScriptCore's bytecode serialisation format & Bun's bundler, now it's 2.6x,” he said.
The package breakdown shows the reduction. On macOS arm64, bytecode fell from 258 MB to 88 MB, a 66% reduction. Runtime components account for 66 MB and the application’s JavaScript for another 34 MB in the latest build.
The change comes as Claude Code continues to ship frequent releases. Version 2.1.248 was published on August 27, while version 2.1.250 followed on August 28.
The latest package remains a terminal-based coding agent that can understand codebases, edit files, execute commands, and handle Git workflows. The smaller binary therefore reduces the software users need to download and install, while the bytecode optimisation is tied specifically to Claude Code’s startup path.
Sumner earlier founded Bun, the JavaScript runtime and all-in-one toolkit Anthropic acquired last December. Bun combines a JavaScript runtime, bundler, package manager, and test runner, and has become core infrastructure for Claude Code.
Anthropic said at the time of the acquisition that Bun would help make Claude Code faster and more stable, while Sumner’s team would continue developing the open-source project.

The size reduction follows a series of recent Claude Code changes aimed at improving runtime efficiency and reliability.
In version 2.1.247, released a day before the size reduction was disclosed, Anthropic fixed unbounded memory growth linked to failed hook or background-task output and stopped excessive error output from wedging sessions with “Prompt is too long” errors.
The release also reduced unnecessary GitHub status checks and extended the effective auto-compaction point for Sonnet 5’s 1-million-token context window from about 934,000 to 967,000 tokens.
- AI Productivity to Add $28 Billion to $84 Billion to India's GDP: Cognizant Study
- OpenAI's Codex Helps in Aerodynamic Engineering for Mazda
- Bengaluru Accounts for 34% of India's FDE Talent: Report
- 84% See AI Productivity Gains, but Only 10% Allow Agent Autonomy: TCS Study
- Snowflake Introduces Business Ontology, Cortex AI Gateway to Bring AI Agents Into Enterprise Workflows
- Three Members Behind Claude Code Leave Anthropic Within 10 Days

