generateText: a first model call

generateText sends a prompt to a model and waits for the whole answer. The result has the text, why the model stopped, and how many tokens it used.

import { generateText } from "ai";
import { MockLanguageModelV4 } from "ai/test";

const model = new MockLanguageModelV4({
  doGenerate: async () => ({
    content: [{ type: "text", text: "Hello from a model." }],
    finishReason: { unified: "stop", raw: undefined },
    usage: {
      inputTokens: { total: 4, noCache: 4, cacheRead: undefined, cacheWrite: undefined },
      outputTokens: { total: 4, text: 4, reasoning: undefined },
    },
    warnings: [],
  }),
});

const result = await generateText({ model, prompt: "Say hello." });
console.log(result.text);
console.log(result.finishReason, result.usage.totalTokens);

npx tsx hello.ts

generateText takes a model and a prompt. It is async, so you await it; tsx allows await at the top level of a file. result.text is the answer, finishReason is why the model stopped, stop meaning it was done, and usage counts tokens.

MockLanguageModelV4 from ai/test is a model whose doGenerate you write. It returns exactly what a provider's model returns to the SDK: a list of content parts, a finish reason and usage. Every provider package implements that same interface, which is why the rest of your code does not care which model it has.