Claude Code v2.1.117 — effort restored, native search, Opus 4.7 context fix

Source: GitHub releases Date: 2026-04-22

Summary

Three significant changes in one release:

1. Default effort restored to high for Pro/Max on Opus 4.6 and Sonnet 4.6 — reversing the medium demotion that triggered community backlash. The effort gate experiment lasted approximately 2 weeks.
2. Native bfs/ugrep replacing Glob/Grep on macOS/Linux native builds — faster searches without tool round-trips. The dependency stack deepens (Claude Code now embeds bfs and ugrep).
3. Opus 4.7 context window fix — was computing against 200K instead of native 1M, causing premature autocompaction and inflated /context percentages. Every Opus 4.7 user was running at 20% of available context.

Also: forked subagents via env var, concurrent MCP connect default, cleanupPeriodDays now sweeps tasks/shell-snapshots/backups, OTEL effort attribute, NO_PROXY fix for Bun.

Implications

The effort restoration is the headline — Anthropic reversed the tier penalty after the Copilot token-billing announcement made the comparison too stark. Restoring effort to high while briefly testing Claude Code removal from Pro suggests the lever they actually want to pull is feature access, not quality degradation. The Opus 4.7 context fix is quietly enormous: every user was effectively running with 200K context for the first week of the model’s life.