Claude Code v2.1.117 brings performance, plugin, and bug fixes
Claude Code `v2.1.117` delivers faster startup, smarter plugin management, expanded OpenTelemetry telemetry, native `bfs`/`ugrep` tools on macOS/Linux, and a large batch of bug fixes including a critical Opus 4.7 context-window miscalculation.
Score breakdown
Developers running Opus 4.7 should update immediately to fix the context-window miscalculation that was triggering premature compaction, and macOS/Linux users gain faster file search with no workflow changes required.
- 01Opus 4.7 context-window bug fixed: Claude Code was computing against 200K instead of the model's native 1M, causing premature autocompaction and inflated `/context` percentages.
- 02On macOS and Linux native builds, `Glob` and `Grep` tools are replaced by embedded `bfs` and `ugrep` via the Bash tool for faster searches without an extra tool round-trip.
- 03Default effort for Pro/Max subscribers on Opus 4.6 and Sonnet 4.6 raised from `medium` to `high`.
Claude Code `v2.1.117`, authored by ashwin-ant, is a substantial release touching performance, agent behavior, plugin infrastructure, observability, and bug fixes. On the performance side, macOS and Linux native builds now replace the `Glob` and `Grep` tools with embedded `bfs` and `ugrep` accessible through the Bash tool, eliminating a separate tool round-trip for file searches. Startup is also faster when both local and claude.ai MCP servers are configured, as concurrent connection is now the default. Windows builds gain cached `where.exe` lookups per process to speed up subprocess launches.
Forked subagents can now be enabled on external builds via the `CLAUDE_CODE_FORK_SUBAGENT=1` environment variable.
Agent and session management received several upgrades. Forked subagents can now be enabled on external builds via the `CLAUDE_CODE_FORK_SUBAGENT=1` environment variable. Agent frontmatter `mcpServers` are now loaded for main-thread agent sessions launched with `--agent`. The `/model` command now persists selections across restarts even when a project pins a different model, and the `/resume` command can summarize stale, large sessions before re-reading them. Default effort for Pro/Max subscribers on Opus 4.6 and Sonnet 4.6 has been raised from `medium` to `high`.
The release also fixes a critical Opus 4.7 bug where Claude Code was computing context against a 200K window instead of the model's native 1M context, causing inflated `/context` percentages and premature autocompaction. Other notable fixes include: OAuth tokens now refreshing reactively on 401 errors instead of dying with "Please run /login"; `WebFetch` no longer hangs on very large HTML pages; HTTP 204 proxy responses now surface a clear error instead of a `TypeError`; and an idle re-render loop causing memory growth on Linux has been resolved. OpenTelemetry observability is expanded with `command_name` and `command_source` fields on `user_prompt` events, and an `effort` attribute on cost, token, and API events when the model supports effort levels.
Key facts
- 01Opus 4.7 context-window bug fixed: Claude Code was computing against 200K instead of the model's native 1M, causing premature autocompaction and inflated `/context` percentages.
- 02On macOS and Linux native builds, `Glob` and `Grep` tools are replaced by embedded `bfs` and `ugrep` via the Bash tool for faster searches without an extra tool round-trip.
- 03Default effort for Pro/Max subscribers on Opus 4.6 and Sonnet 4.6 raised from `medium` to `high`.
- 04Forked subagents can now be enabled on external builds by setting `CLAUDE_CODE_FORK_SUBAGENT=1`.