zhipu-ai-sdk-provider weekly downloads statistics

zhipu-ai-sdk-provider

Zhipu AI provider for the Vercel AI SDK. Integrates GLM language models, embedding models, image generation, and text-to-speech from bigmodel.cn.

Installation

npm install zhipu-ai-sdk-provider ai

Configuration

Set your API key from bigmodel.cn:

export ZHIPU_API_KEY=your-api-key

Quick Start

import { generateText } from "ai";
import { zhipu } from "zhipu-ai-sdk-provider";

const { text } = await generateText({
  model: zhipu("glm-4.7"),
  prompt: "Why is the sky blue?",
});

console.log(text);

Provider Setup

Default instance — reads ZHIPU_API_KEY from environment:

import { zhipu } from "zhipu-ai-sdk-provider";

Custom instance — pass your own config:

import { createZhipu } from "zhipu-ai-sdk-provider";

const zhipu = createZhipu({
  apiKey: "your-api-key",
  baseURL: "https://open.bigmodel.cn/api/paas/v4", // default
  headers: { "X-Custom": "value" }, // optional
});

Settings

Option	Type	Description
apiKey	string	API key. Defaults to ZHIPU_API_KEY env var.
baseURL	string	API base URL. Defaults to https://open.bigmodel.cn/api/paas/v4.
headers	Record<string, string>	Custom headers to include in requests.
fetch	FetchFunction	Custom fetch implementation (e.g. for testing or proxying).

Supported Models

Language Models

Model	Description
glm-5	Latest flagship, designed for agent applications
glm-4.7	High-performance with deep thinking
glm-4.7-flash	Fast variant of GLM-4.7
glm-4.7-flashx	Extended fast variant of GLM-4.7
glm-4.6	Previous-gen flagship, supports tool_stream
glm-4.5	GLM-4.5 with thinking support
glm-4.5-air	Lightweight GLM-4.5
glm-4.5-x	Extended GLM-4.5
glm-4.5-airx	Extended lightweight GLM-4.5
glm-4.5-flash	Fast GLM-4.5
glm-4-plus	Enhanced GLM-4
glm-4-air	Lightweight GLM-4
glm-4-air-250414	GLM-4-air dated snapshot
glm-4-airx	Extended lightweight GLM-4
glm-4-long	Long-context GLM-4
glm-4-flash	Fast GLM-4
glm-4-flash-250414	GLM-4-flash dated snapshot
glm-4-flashx	Extended fast GLM-4
glm-4-32b-0414-128k	Open-source 32B model, 128K context

Vision Models

Model	Description
glm-4.6v	GLM-4.6 vision model
glm-4.6v-flash	Fast GLM-4.6 vision
glm-4.6v-flashx	Extended fast GLM-4.6 vision
glm-4.5v	GLM-4.5 vision model
glm-4v-plus-0111	Enhanced GLM-4 vision
glm-4v-plus	GLM-4 vision
glm-4v	Standard GLM-4 vision
glm-4v-flash	Fast GLM-4 vision

Reasoning Models

Model	Description
glm-z1-air	Lightweight reasoning model
glm-z1-airx	Extended reasoning model
glm-z1-flash	Fast reasoning model
glm-4.1v-thinking-flash	Vision + reasoning
glm-4.1v-thinking-flashx	Extended vision + reasoning

Embedding Models

Model	Description
embedding-3	Latest embedding model (recommended)
embedding-2	Standard embedding model

Image Generation Models

Model	Description
cogview-4-250304	Latest CogView 4, supports quality: "hd"
cogview-4	Standard CogView 4
cogview-3-flash	Fast image generation

Speech Models

Model	Description
glm-tts	Text-to-speech model

Usage

Text Generation

import { generateText } from "ai";
import { zhipu } from "zhipu-ai-sdk-provider";

const { text } = await generateText({
  model: zhipu("glm-4.7"),
  prompt: "Explain quantum computing in simple terms.",
});

Streaming

import { streamText } from "ai";
import { zhipu } from "zhipu-ai-sdk-provider";

const result = streamText({
  model: zhipu("glm-4.7"),
  prompt: "Write a short story about a robot.",
});

for await (const chunk of result.textStream) {
  process.stdout.write(chunk);
}

Thinking / Reasoning Mode

GLM-5 and GLM-4.7 models have thinking enabled by default. You can control this via model settings:

// Enable with preserved thinking (retains reasoning across turns)
const { text } = await generateText({
  model: zhipu("glm-4.7", {
    thinking: {
      type: "enabled",
      clearThinking: false, // preserve reasoning context across turns
    },
  }),
  prompt: "Solve step by step: What is 23 * 47?",
});

// Disable thinking for faster responses
const { text } = await generateText({
  model: zhipu("glm-4.7", {
    thinking: { type: "disabled" },
  }),
  prompt: "What is 2 + 2?",
});

You can also configure thinking at call time with zhipuOptions:

import { generateText } from "ai";
import { zhipu, zhipuOptions } from "zhipu-ai-sdk-provider";

const { text } = await generateText({
  model: zhipu("glm-4.7"),
  prompt: "Solve this complex math problem.",
  providerOptions: zhipuOptions({
    thinking: { type: "enabled", clear_thinking: false },
  }),
});

Provider Options (zhipuOptions)

Use zhipuOptions to pass Zhipu-specific API parameters at call time. Options are passed directly — no nesting required:

import { generateText } from "ai";
import { zhipu, zhipuOptions } from "zhipu-ai-sdk-provider";

const { text } = await generateText({
  model: zhipu("glm-4.7"),
  prompt: "Hello!",
  providerOptions: zhipuOptions({
    temperature: 0.7,
    top_p: 0.9,
    max_tokens: 4096,
    do_sample: true,
  }),
});

Available Options…