CLAUDE CODE 2.1.74 STOPS NODE STREAMING MEMORY LEAKS AND ADDS ENTERPRISE-GRADE MODEL ROUTING
Anthropic shipped Claude Code 2.1.73–2.1.74 with a key Node.js memory leak fix, better provider routing, and sturdier enterprise auth. The 2.1.74 release fixes...
Anthropic shipped Claude Code 2.1.73–2.1.74 with a key Node.js memory leak fix, better provider routing, and sturdier enterprise auth.
The 2.1.74 release fixes a streaming memory leak on the Node.js path when a generation is canceled early, adds actionable tips to the /context command, and introduces an autoMemoryDirectory setting for controlling on-disk memory v2.1.74. It also tightens OAuth flows, resolves model ID handling in agent configs, and addresses Windows LSP and terminal RTL glitches.
The prior 2.1.73 release added modelOverrides so the model picker can map cleanly to custom provider IDs like Bedrock inference profile ARNs, and hardened behavior under corporate proxies and complex shell prompts v2.1.73. Together, these updates reduce resource blowups, simplify multi-cloud routing, and remove several enterprise blockers.
If you run long streams, work behind a proxy, or route across Bedrock, Vertex, or Foundry, upgrade and retest with 2.1.74.
The Node.js streaming memory leak fix prevents unbounded RSS growth that can slow services or trigger OOM kills.
Model routing controls and auth fixes unblock cleaner multi-cloud deployments and reduce flaky enterprise login issues.
-
terminal
Run long generation streams and early-cancel scenarios on the Node.js path to verify RSS stabilizes and no buffers linger.
-
terminal
Validate OAuth behind corporate proxies and expiring refresh tokens, and confirm modelOverrides correctly map to Bedrock ARNs.
Legacy codebase integration strategies...
- 01.
Upgrade to 2.1.74 and audit hooks, plugin overrides (--plugin-dir), and OAuth callbacks/refresh across environments.
- 02.
On Windows, verify LSP servers and terminal rendering for RTL languages; on macOS, confirm voice mode prompts appear.
Fresh architecture paradigms...
- 01.
Standardize agent configs with modelOverrides and explicit full model IDs across Bedrock, Vertex, and Foundry.
- 02.
Set autoMemoryDirectory and use /context suggestions to control context size, memory bloat, and tool usage.