Claude Code 2.1.323 Ships Experimental Code Mode
Anthropic released Claude Code version 2.1.323 on November 12, 2026, according to the project's official GitHub release entry GitHub. The update introduces Code Mode, a new experimental feature that changes how the coding assistant interacts with its own tools.
What Code Mode Changes
In the release notes, Anthropic describes Code Mode as "an experimental opt-in in which Claude writes JavaScript to call its tools" GitHub. The company's documentation explains that this allows a single JavaScript program to make multiple tool calls, including parallel calls, and to pass the result of one call directly into a subsequent one Claude Code docs.
That marks a departure from how Claude Code has worked until now. The documentation contrasts Code Mode with the standard mode, in which Claude requests tools individually and must wait for each result before deciding on its next action Claude Code docs. By letting Claude write orchestration logic upfront, Code Mode is designed to cut down on the back-and-forth of turn-by-turn tool requests that has characterized agentic coding sessions.
How to Turn It On
Code Mode is off by default. Developers who want to try it must set the environment variable `CLAUDE_CODE_EXPERIMENTAL_CODE_MODE=1`, per Anthropic's documentation Claude Code docs. Anyone running an older build of Claude Code first needs to update to version 2.1.323 or later before the flag will have any effect, since the feature does not exist in prior releases.
The experimental label signals that Anthropic is treating this as a trial feature rather than a committed direction for the product. No rollout timeline for a stable, default-on version has been published alongside this release.
Why It Matters for Agent Workflows
Tool-calling overhead has been a persistent bottleneck in agentic coding tools: each individual tool request and response round-trip adds latency and consumes context. By letting Claude express a sequence of tool calls as a single script, Code Mode is aimed squarely at developers building complex, multi-step coding agents who want fewer interruptions between planning and execution. Because the feature is opt-in and experimental, its real-world performance and reliability compared with the standard mode have not yet been independently benchmarked.
The change arrives as part of Claude Code's regular release cadence, with version 2.1.323 otherwise following Anthropic's standard update and documentation process for the coding tool.


