Release Notes v0.1.23

New Features

• Thinking Token Budget Support — Models that perform internal reasoning (o1, o3, Claude, Gemini thinking) can now reserve tokens for thinking separately from output. A new thinkingTokenBudget setting is available in Options, with validation to ensure it stays within reasonable bounds.

Improvements

• Suggestion Output Tokens Increased — Maximum output tokens for suggestions raised from 1024 to 4096, allowing more comprehensive suggestions.
• Streaming Token Quality — Streamed tokens are now post-processed as they're appended, stripping model thinking blocks, orphaned tags, and XML processing instructions to prevent broken tokens or formatting artifacts in the editor.
• Token Calculation Refinements — Improved computation of effective max tokens when thinking is supported, and better handling of user-configured thinking budgets.

Bug Fixes

• Guard added for empty partial suggestions to prevent errors.
• Type safety improvements around thinking token handling in suggestion and improvement utilities.