You can now adjust the model thinking level (off, low, medium, high, xhigh, max) directly from the Task Chat model selector via a smooth discrete slider, with automatic clamping when switching between Auto, Claude, GPT, Gemini, Grok, and DeepSeek models.
Bots & Scheduled Timers
You can now configure and persist a dedicated thinking level on any Bot or Scheduled Timer, or inherit the source conversation's thinking level when creating a timer from chat.
Kollab CLI & Bundled Skills
You can now pass --thinking <level> across kollab model run, kollab task create/ask/update/answer-question, kollab bot create/update, and kollab timer create/update, and bundled skills automatically inherit the active session thinking level.
LLM Gateway
High-balance workspaces now support up to 20 concurrent in-flight requests per external API key on the LLM Gateway, and output_config.effort is normalized consistently across all supported model families.
