A custom provider extension that exposes ZAI-family models from two upstream hosts:

• Cerebras-hosted models (requires CEREBRAS_API_KEY)
• ZAI-hosted models (requires ZAI_API_KEY)

Model availability is determined strictly by which provider API keys are present.

• Key-based model availability
  • CEREBRAS_API_KEY => Cerebras model set
  • ZAI_API_KEY => ZAI model set
  • Both keys => both model sets
• Model-driven endpoint routing
  • zai-custom/<model-id> is enough to choose the right base URL and API key
• Reasoning + sampling knobs
  • Supports temperature, top_p, and clear_thinking

When invoking the provider, you can pass these options:

• temperature: Sampling temperature (0.0-2.0)
• top_p: Nucleus sampling parameter
• clear_thinking: Whether to clear thinking output

The provider is registered as zai-custom.

export CEREBRAS_API_KEY="your-cerebras-key" export ZAI_API_KEY="your-zai-key"

GLM-4.7 Cerebras (zai-glm-4.7)

• Endpoint: https://api.cerebras.ai/v1
• Reasoning: false
• Context Window: 131,072 tokens
• Max Output: 40,000 tokens
• Input Cost: $0.00 / 1M tokens
• Output Cost: $0.00 / 1M tokens
• Cache Read: $0.00 / 1M tokens

GLM 4.7 ZAI (glm-4.7)

• Endpoint: https://api.z.ai/api/coding/paas/v4
• Reasoning: true
• Context Window: 204,800 tokens
• Max Output: 131,072 tokens
• Input Cost: $0.60 / 1M tokens
• Output Cost: $2.20 / 1M tokens
• Cache Read: $0.11 / 1M tokens

GLM-5 (ZAI) (glm-5)

• Endpoint: https://api.z.ai/api/coding/paas/v4
• Reasoning: true
• Context Window: 200,000 tokens
• Max Output: 128,000 tokens
• Input Cost: $0.15 / 1M tokens
• Output Cost: $0.60 / 1M tokens
• Cache Read: $0.00 / 1M tokens