Reasoning

node-llm provides a unified way to access the “thinking” or “reasoning” process of models like DeepSeek R1 or OpenAI o1/o3. Some models expose their internal chain of thought before providing the final answer.

Accessing Reasoning Content

For models that support it (like deepseek-reasoner), you can access the reasoning text via the .reasoning property on the response object.

import { LLM } from "@node-llm/core";

const chat = LLM.chat("deepseek-reasoner");
const response = await chat.ask("Prove that the square root of 2 is irrational.");

// Show the inner thought process
console.log("Thinking:", response.reasoning);

// Show the final answer
console.log("Answer:", response.content);

Streaming Reasoning

When using .stream(), reasoning content is emitted as chunks, just like regular content. You can distinguish them by checking chunk.reasoning.

const chat = LLM.chat("deepseek-reasoner");

for await (const chunk of chat.stream("Explain quantum entanglement")) {
  if (chunk.reasoning) {
    process.stdout.write(`[Thinking] ${chunk.reasoning}`);
  }
  if (chunk.content) {
    process.stdout.write(chunk.content);
  }
}

The final response object in the onEndMessage callback or returned by the stream will also contain the full aggregated reasoning string.

OpenAI o1/o3 Support

OpenAI models like o3-mini do not expose the reasoning text directly (in a separate field), but they use “reasoning tokens” during generation. node-llm automatically tracks these tokens and includes them in the usage and cost calculations.

const chat = LLM.chat("o3-mini");
const response = await chat.ask("Write a complex algorithm");

console.log(`Reasoning tokens used: ${response.usage.output_tokens}`);
console.log(`Total cost: $${response.cost}`);

Supported Models

Currently, the following models have enhanced reasoning support in node-llm:

Model ID	Provider	Feature
deepseek-reasoner	DeepSeek	Full thinking text extraction
o1-preview, o1-mini, o3-mini	OpenAI	Reasoning token & cost tracking