GLM (Zhipu AI)

Access Zhipu GLM models and a family of Chinese AI providers through a unified configuration. Includes aliases for Minimax, Moonshot (Kimi), Qwen (DashScope), and Z.AI.

Prerequisites

• A Zhipu AI API key from open.bigmodel.cn (for GLM models), or
• API keys for the specific provider you want to use (Minimax, Moonshot, Qwen, etc.)

Quick Setup

1. Get API Key

1. Sign up at open.bigmodel.cn
2. Navigate to the API Keys section
3. Create a new key (format: id.secret)

2. Configure

[default]
provider = "glm"
model = "glm-4-plus"

[providers.glm]
api_key = "${GLM_API_KEY}"

Or set the environment variable:

export GLM_API_KEY="abc123.secretXYZ"

3. Verify

prx doctor models

Available Models

GLM Models

Model	Context	Vision	Tool Use	Notes
glm-4-plus	128K	Yes	Yes	Most capable GLM model
glm-4	128K	Yes	Yes	Standard GLM-4
glm-4-flash	128K	Yes	Yes	Fast and cost-effective
glm-4v	128K	Yes	Yes	Vision-optimized

Aliased Providers

PRX also supports these providers as aliases that route through the OpenAI-compatible interface:

Provider	Alias Names	Base URL	Key Models
Minimax	minimax, minimax-intl, minimax-cn	api.minimax.io/v1 (intl), api.minimaxi.com/v1 (CN)	MiniMax-Text-01, abab6.5s
Moonshot	moonshot, kimi, moonshot-intl, kimi-cn	api.moonshot.ai/v1 (intl), api.moonshot.cn/v1 (CN)	moonshot-v1-128k, moonshot-v1-32k
Qwen	qwen, dashscope, qwen-intl, qwen-us	dashscope.aliyuncs.com (CN), dashscope-intl.aliyuncs.com (intl)	qwen-max, qwen-plus, qwen-turbo
Z.AI	zai, z.ai, zai-cn	api.z.ai/api/coding/paas/v4 (global), open.bigmodel.cn/api/coding/paas/v4 (CN)	Z.AI coding models

Configuration Reference

GLM (Native Provider)

Field	Type	Default	Description
api_key	string	required	GLM API key in id.secret format
model	string	required	GLM model name

Aliased Providers (OpenAI-Compatible)

Field	Type	Default	Description
api_key	string	required	Provider-specific API key
api_url	string	auto-detected	Override the default base URL
model	string	required	Model name

Features

JWT Authentication

GLM uses JWT-based authentication rather than plain API keys. PRX automatically:

1. Splits the API key into id and secret components
2. Generates a JWT token with:
   • Header: {"alg":"HS256","typ":"JWT","sign_type":"SIGN"}
   • Payload: {"api_key":"<id>","exp":<expiry_ms>,"timestamp":<now_ms>}
   • Signature: HMAC-SHA256 with the secret key
3. Caches the JWT for 3 minutes (token expires at 3.5 minutes)
4. Sends it as Authorization: Bearer <jwt>

Regional Endpoints

Most aliased providers offer both international and China-mainland endpoints:

# International (default for most)
provider = "moonshot-intl"

# China mainland
provider = "moonshot-cn"

# Explicit regional variants
provider = "qwen-us"      # US region
provider = "qwen-intl"    # International
provider = "qwen-cn"      # China mainland

Minimax OAuth Support

Minimax supports OAuth token authentication:

export MINIMAX_OAUTH_TOKEN="..."
export MINIMAX_OAUTH_REFRESH_TOKEN="..."

Set provider = "minimax-oauth" or provider = "minimax-oauth-cn" to use OAuth instead of API key authentication.

Qwen OAuth and Coding Modes

Qwen offers additional access modes:

• Qwen OAuth: provider = "qwen-oauth" or provider = "qwen-code" for OAuth-based access
• Qwen Coding: provider = "qwen-coding" or provider = "dashscope-coding" for the coding-specialized API endpoint

Provider Aliases Reference

Alias	Resolves To	Endpoint
glm, zhipu, glm-global, zhipu-global	GLM (global)	api.z.ai/api/paas/v4
glm-cn, zhipu-cn, bigmodel	GLM (CN)	open.bigmodel.cn/api/paas/v4
minimax, minimax-intl, minimax-global	MiniMax (intl)	api.minimax.io/v1
minimax-cn, minimaxi	MiniMax (CN)	api.minimaxi.com/v1
moonshot, kimi, moonshot-cn, kimi-cn	Moonshot (CN)	api.moonshot.cn/v1
moonshot-intl, kimi-intl, kimi-global	Moonshot (intl)	api.moonshot.ai/v1
qwen, dashscope, qwen-cn	Qwen (CN)	dashscope.aliyuncs.com
qwen-intl, dashscope-intl	Qwen (intl)	dashscope-intl.aliyuncs.com
qwen-us, dashscope-us	Qwen (US)	dashscope-us.aliyuncs.com
zai, z.ai	Z.AI (global)	api.z.ai/api/coding/paas/v4
zai-cn, z.ai-cn	Z.AI (CN)	open.bigmodel.cn/api/coding/paas/v4

Troubleshooting

"GLM API key not set or invalid format"

The GLM API key must be in id.secret format (contains exactly one dot). Verify your key format:

abc123.secretXYZ  # correct
abc123secretXYZ   # wrong - missing dot

JWT generation fails

Ensure your system clock is accurate. JWT tokens include a timestamp and expire after 3.5 minutes.

MiniMax "role: system" rejected

MiniMax does not accept role: system messages. PRX automatically merges system message content into the first user message when using MiniMax providers.

Qwen/DashScope timeout

Qwen's DashScope API requires HTTP/1.1 (not HTTP/2). PRX automatically forces HTTP/1.1 for DashScope endpoints. If you experience timeouts, ensure your network allows HTTP/1.1 connections.

Regional endpoint errors

If you get connection errors, try switching between regional endpoints:

• China users: Use *-cn variants
• International users: Use *-intl or base variants
• US-based users: Try qwen-us for Qwen