Open Models Almanac
Methodsinference & servingagentic scaffolding

single source · inference & serving

Assistant output channels

Assistant output channels tag every message the model emits as analysis, commentary, or final.1 The channels separate internal messages, such as tool calls routed to the analysis channel, from user-facing output.2,3

How this definition was written

Summarised by glm-5.3-flash from 1 source; every sentence is checked against the quote it cites.

  1. Valid channels: analysis, commentary, final. Channel must be included for every message — OpenAI Harmony Response Format
  2. Requests will be made to the analysis channel — OpenAI Harmony Response Format
  3. The code will not be shown to the user — OpenAI Harmony Response Format

Also called channels.

Filed under inference & serving › agentic scaffolding