-
New model picker UX. Each provider's recommended model and recommended thinking level now show a "*" badge in the chip dropdowns. A brief explainer line ("Best for | Cost | Speed | Note") appears below the chips when you change the selection and fades after 10 seconds. Hovering any row in the thinking dropdown previews that combo's explainer before you click. v1.1.4 wipes any existing model + thinking-level picks once on first launch so everyone lands on the new defaults; re-pick from the chip dropdowns if you preferred a different combo.
-
New default model + thinking picks for built-in providers, based on bench testing of multi-step REAPER scripting tasks. Anthropic: Sonnet 4.6 (no thinking) remains the recommended Claude default; Haiku 4.5 default thinking moves Low to High. OpenAI: full GPT-5.4 (no thinking) replaces mini as the recommended OpenAI default; nano default thinking moves Low to None. Gemini: Flash Lite default thinking moves Minimal to Low; Flash 3 default thinking moves Low to Minimal.
-
Help page Quick model guide added under Providers & Models with the recommended pick per provider. Help page renderer now handles markdown tables and `###` sub-headings, so the existing Stock Fallbacks table and the new Quick model guide render properly instead of as raw markdown.
-
ReaComp Threshold conversion clarified in the plugin reference (`value = 10^(dB / 20)` formula + worked examples), reducing the rate at which lighter models pick the default normalized value when asked for a specific dB target.