Choose reasoning effort - OpenServ Docs

Three calibrated reasoning paths balancing depth, speed, and cost

Reasoning-capable models accept low, medium, or high effort.

const response = await client.chat.completions.create({
  model: "gpt-5.4-mini",
  reasoning_effort: "medium",
  messages: [
    { role: "system", content: "Solve constrained planning problems carefully." },
    { role: "user", content: "Plan this project under these constraints: ..." },
  ],
});

Start with low. Move higher only when evaluation shows a measurable improvement that justifies extra cost and latency.