The contract for agentic terminal work, where Astra's lead is clearest: 57.9 percent on Terminal-Bench 4.0 against 37.3 for GPT-5.6 Sol and 55.8 for Claude Fable 5.1, at lower estimated cost per task in OpenAI's figures. Reasoning: medium for agentic coding, which is where OpenAI's own mapping puts it; high when the task is complex debugging or deep planning. API reasoning.effort "medium" or "high"; Codex Astra Medium, or Extra High for the hard ones. API users: tool calling needs the Responses API, and Astra does not accept temperature or top_p. The conventions lines are OpenAI's own developer message from its coding evals. Free template — copy and paste into ChatGPT, Claude, or Copilot.