Z.AI Provider

The Z.AI provider gives you access to GLM language and vision models through the Z.AI API.

API keys can be created in the Z.AI API key console.

Setup

The Z.AI provider is available through the @ai-sdk/zai package. Install it with:

pnpm add @ai-sdk/zai

Set the ZAI_API_KEY environment variable:

1
ZAI_API_KEY=your-api-key

Provider Instance

Import the default provider instance zai from @ai-sdk/zai:

1
import { zai } from '@ai-sdk/zai';

For custom configuration, use createZai:

1
import { createZai } from '@ai-sdk/zai';
2

3
const zai = createZai({
4
  apiKey: process.env.ZAI_API_KEY,
5
  baseURL: 'https://api.z.ai/api/paas/v4',
6
});

The provider accepts these optional settings:

• apiKey string

  API key sent in the Authorization header. It defaults to the ZAI_API_KEY environment variable.

• baseURL string

  URL prefix for API requests. It defaults to https://api.z.ai/api/paas/v4.

• headers Record<string, string>

  Additional request headers.

• fetch FetchFunction

  Custom fetch implementation, for example to intercept requests in middleware.

Language Models

Create a language model by passing its model id to the provider:

1
import { zai } from '@ai-sdk/zai';
2
import { generateText } from 'ai';
3

4
const { text } = await generateText({
5
  model: zai('glm-5.3'),
6
  prompt: 'Explain quantum entanglement in simple terms.',
7
});

Current model families include GLM 5, GLM 4.7, GLM 4.6, GLM 4.5, and the GLM vision models. Model availability changes over time; see the Z.AI model documentation for the current catalog.

The provider supports:

• text generation and streaming
• reasoning output and reasoning history
• function calling, including incremental tool-call streaming
• JSON object output
• URL-based image and video inputs on compatible vision models

Provider Options

Z.AI-specific options can be passed through providerOptions.zai:

1
import { zai } from '@ai-sdk/zai';
2
import { generateText } from 'ai';
3

4
const result = await generateText({
5
  model: zai('glm-5.3'),
6
  prompt: 'Compare two approaches to implementing a rate limiter.',
7
  providerOptions: {
8
    zai: {
9
      thinking: { type: 'enabled', clearThinking: true },
10
      reasoningEffort: 'high',
11
      requestId: 'request-123456',
12
      userId: 'user-123456',
13
    },
14
  },
15
});

The following options are available:

• doSample boolean

  Enables sampling. When disabled, temperature and topP do not take effect.

• thinking object

  Controls thinking with type: 'enabled' | 'disabled'. Set clearThinking to false to retain reasoning from previous assistant messages.

• reasoningEffort 'none' | 'minimal' | 'low' | 'medium' | 'high' | 'xhigh' | 'max'

  Controls reasoning effort on supported models.

• toolStream boolean

  Enables incremental function-call argument streaming on supported models.

• requestId string

  A caller-provided request id between 6 and 64 characters.

• userId string

  A non-sensitive end-user id between 6 and 128 characters.

Streaming Tool Calls

Enable toolStream when you want supported GLM models to stream function-call arguments incrementally:

1
import { zai } from '@ai-sdk/zai';
2
import { streamText, tool } from 'ai';
3
import { z } from 'zod';
4

5
const result = streamText({
6
  model: zai('glm-5.3'),
7
  prompt: 'What is the weather in San Francisco?',
8
  tools: {
9
    weather: tool({
10
      description: 'Get the weather for a city',
11
      inputSchema: z.object({ city: z.string() }),
12
    }),
13
  },
14
  providerOptions: {
15
    zai: { toolStream: true },
16
  },
17
});
18

19
for await (const part of result.fullStream) {
20
  console.log(part);
21
}