Claude Code Restores High Effort Default for Pro/Max: Reversing the March Downgrade

Claude Code Restores high as the Default Effort for Pro and Max

Starting with version 2.1.117, Claude Code raises the default effort level back to high for Pro and Max subscribers using Opus 4.6 and Sonnet 4.6. The previous default was medium, a setting that Anthropic had quietly introduced in March 2026 with version 2.1.68 — and which generated one of the most vocal user backlash episodes in Claude Code's history.

The March Downgrade and Community Reaction

In early March 2026, Anthropic changed the default effort level from high to medium for Max and Team plan subscribers. The stated rationale was that medium offered a better speed-and-cost balance for most coding tasks. However, many users experienced a noticeable quality drop without understanding why — the change appeared in the changelog but was missed by a large portion of the user base, and the brief in-app dialog did not register for many developers.

The reaction was sharp. An AMD senior director published a widely circulated analysis of 6,852 Claude Code session files, 17,871 thinking blocks, and 234,760 tool calls, arguing that "Claude has regressed to the point it cannot be trusted to perform complex engineering." The post triggered coverage from VentureBeat, Fortune, and Axios, with Anthropic's head of Claude Code, Boris Cherny, responding publicly on X to explain the rationale and confirm that users could override the setting manually.

What Changes in v2.1.117

Version 2.1.117 reverses the default. Pro and Max subscribers on Opus 4.6 and Sonnet 4.6 now start sessions at high effort — the same default that existed before the March downgrade. Users who had already manually set their effort level will see no change; the update only affects the out-of-the-box default.

The effort system in Claude Code controls how much adaptive reasoning the model applies. The available levels — low, medium, high, xhigh, and max — let users tune the tradeoff between speed and depth. xhigh remains the default for Opus 4.7 (introduced with v2.1.111), and max is a session-only override that is not persisted across restarts.

Opus 4.7 Context Window Fix

The release also closes a significant bug for Opus 4.7 users: the /context command was displaying inflated usage percentages because Claude Code was computing token consumption against a 200K context window rather than Opus 4.7's native 1M context window. The consequence was that sessions appeared to be approaching capacity far earlier than they actually were, triggering premature autocompaction and causing users to lose conversation history unnecessarily. This fix ensures that context tracking is accurate for Opus 4.7's full 1M window.

Other Notable Changes

Several additional improvements ship in v2.1.117:

• The /resume command now offers to summarize stale, large sessions before re-reading them — consistent with the existing --resume CLI flag behavior. This reduces cold-start latency when returning to old conversations.
• On macOS and Linux, the built-in Glob and Grep tools are replaced by embedded bfs and ugrep binaries, available through the Bash tool. The result is faster file search without an extra tool round-trip. Windows and npm-installed builds are not affected by this change.
• MCP startup is faster when both local and claude.ai MCP servers are configured, as concurrent connection is now the default.
• A long-standing OAuth session bug is fixed: Plain-CLI sessions no longer die mid-session with "Please run /login" when the access token expires — the token is now refreshed reactively on 401.
• Fixed WebFetch hanging on very large HTML pages by truncating input before conversion.
• Fixed NO_PROXY not being respected for remote API requests when running under Bun.
• Fixed rare spurious escape/return triggers over slow connections.