Two releases in three days, both with substance: Claude Code 2.1.116 (April 20) and 2.1.117 (April 22) bring a handful of improvements you’ll actually notice in daily use.
/resume Finally Gets Fast
The biggest pain point for power users has always been resuming large sessions. With 2.1.116, /resume on sessions over 40 MB now loads up to 67 percent faster. If you regularly work with large codebases and leave sessions running overnight, you’ll feel the difference immediately.
There’s also a thinking spinner that now shows inline progress: “still thinking,” “thinking more,” “almost done thinking.” Sounds like a detail, but when Claude thinks for 30 seconds, it’s reassuring to see something is happening.
Native Binaries — A Quiet Architecture Shift
With 2.1.117, Claude Code switches Glob and Grep to embedded native tools: bfs replaces the previous Glob implementation, ugrep replaces the previous Grep solution. This means faster file searches in large repos — especially on machines where Node.js-based file operations were the bottleneck.
The CLI now also spawns a native binary instead of bundled JavaScript. This reduces startup time and memory usage.
Opus 4.7 Context Window Fix
A bug many users probably noticed: Claude Code was miscalculating the Opus 4.7 context window — using 200K instead of the actual 1M tokens. That’s now fixed. If you work with Opus 4.7, you finally get the full context.
Other Highlights
MCP servers now start in parallel instead of sequentially — if you have multiple MCP servers configured, you save several seconds on every startup. Forked subagents can be enabled on external builds via an environment variable (CLAUDE_CODE_FORK_SUBAGENT=1). And /resume now offers to create a summary for stale, large sessions instead of loading the entire context.
Plus the usual bug fixes: OAuth token refresh on plain CLI, broken keyboard shortcuts (Ctrl+Z, Cmd+Left/Right), and a security fix for sandbox auto-allow on critical directories.
Sources: