Version 2.1.107

A new system-level optimization for Opus 4.6 that tunes the model's extended thinking behavior, reducing unnecessary reasoning overhead on simple follow-up messages while preserving full thinking on complex tasks.

• Adds a thinking_guidance section to the system prompt instructing Opus 4.6 to calibrate thinking frequency based on task complexity
• On follow-up turns (after the first assistant response), appends a <system-reminder> tag hinting the model to skip thinking blocks for straightforward actions
• Only activates when: using Opus 4.6, thinking is not disabled, and no custom system prompt is set
• Gated by server-side flag loud_sugary_rock — rolling out gradually

This should result in noticeably faster responses on simple follow-up requests (e.g., "run the tests", "fix that typo") while maintaining full reasoning depth for complex tasks that need it.

Thinking frequency tuning system (search for "tune your thinking frequency") — gated by loud_sugary_rock in clientDataCache, system reminder at meK (search for "without a thinking block")

Subagents using Opus 4.6 now automatically use the 1M context window when the feature is available, matching the behavior already offered in main conversation model selection.

• When 1M context is available and the resolved subagent model is Opus 4.6, the [1m] suffix is automatically appended
• Applied in both the explicit model override and default model code paths of subagent model resolution
• Does not apply if the model string already includes [1m]

Auto-upgrade function nc4 (search for "opus-4-6" near "[1m]") applied in subagent model resolution function Eh6

The progress messages shown during long-running operations now appear much sooner, reducing the time users wait before seeing reassurance that the model is still working.

• "Thinking a bit longer… still working on it…" — now at 10s (was 30s)
• "Hang tight… really working through this one…" — now at 30s (was 60s)
• "This is a harder one… it might take another minute…" — now at 50s (was 90s)
• "Still going… thanks for hanging in there…" — now at 80s (was 150s)
• "Taking the time to get this right… thanks for your patience…" — now at 120s (was 240s)

The first progress indicator now appears 3x faster than before (10s vs 30s), providing much quicker feedback during complex reasoning.

Timing thresholds in progress message array (search for "Thinking a bit longer")