The three-tier reasoning recipe: max for planning, mid for execution, high for review
卡尔的AI沃茨 · wechat · 2026-10-12
A Chinese AI blogger shares a practical workflow that assigns model reasoning effort by task stage:
- Planning (max reasoning): use plan mode with voice/image/video input to dump unfiltered ideas — high reasoning organizes fragments into a complete plan, avoiding information loss from typing.
- Execution (mid reasoning): for coding/agent work, mid tier matches quality at much higher speed, provided the requirement spec is detailed enough — citing TerminalBench3.0 results that detailed spec + mid beats vague spec + high.
- Review (high reasoning): have the model audit its own mid-tier output; the author estimates it catches 20% of issues, at about one-third the cost of generating from scratch.
Always-max fails because it's slow and overthinks during execution; a clear plan plus mid-tier iteration is more stable.
More from coding & agent
- Hallmark v2 questioned: how does it check taste without a human looking? — petrusenko_max · 2026-10-23
- Laude Institute launches FrontierCS 2: agents must beat deleted paper contributions blind — lateinteraction · 2026-10-23
- Index Network open-sources Social Discovery Protocol: agents negotiate fit on your behalf — carsonfarmer · 2026-10-23
- Installed a formal-verification skill: it silently installed Java, MacBook runs 10 degrees hotter — zealcaiden · 2026-10-23
- OPAL: an LLM agent rewrites the optimizer's source code during training — RobertTLange · 2026-10-23
- OpenAI's Decisions API hits GA: near real-time routing up to 10x faster than GPT-6 Luna — OpenAIDevs · 2026-10-23