Control DeepSeek reasoning

Enable thinking

Set thinking.type explicitly so behavior does not depend on a host default. Gonka24 also accepts the native enable_thinking switch. Both forms are normalized and forwarded to every compatible upstream.

const response = await client.chat.completions.create({
  model: "deepseek-v4-flash-0731",
  messages: [{ role: "user", content: "Solve this carefully." }],
  thinking: { type: "enabled" },
  reasoning_effort: "high",
});

console.log(response.choices[0]?.message.content);

Equivalent native switch:

const response = await client.chat.completions.create({
  model: "deepseek-v4-flash-0731",
  messages: [{ role: "user", content: "Solve this carefully." }],
  enable_thinking: true,
  reasoning_effort: "high",
});

Choose a mode

Gonka24 supports three public reasoning modes: off, low, and high. Use the thinking switch—not an off effort value—to disable reasoning.

DeepSeek ignores temperature, top_p, presence_penalty, and frequency_penalty while thinking mode is enabled.

Disable thinking

For short or latency-sensitive requests, disable thinking explicitly with thinking.type or enable_thinking: false.

const response = await client.chat.completions.create({
  model: "deepseek-v4-flash-0731",
  messages: [{ role: "user", content: "Answer briefly." }],
  thinking: { type: "disabled" },
});

Read reasoning

Gonka24 preserves both choices[0].message.reasoning and its compatible alias reasoning_content when the selected upstream returns a reasoning trace. Read the final answer from choices[0].message.content.

OpenBroker compatibility tests confirm that enable_thinking: true returns the trace separately. Gonka24 maps the official thinking.type switch to that native flag. reasoning_tokens is preserved when supplied by an upstream, but is not estimated when absent.

See the official DeepSeek thinking-mode guide for the upstream contract.