GLM family

Zhipu’s GLM text models, including the GLM-5.3, GLM-5.2, GLM-5.1 and GLM-5 series and the vision model GLM-5V Turbo. They are called through the OpenAI-compatible API and also support the Anthropic-native API, so they work in Claude Code.

Common tasks

Models

For the full model list and live prices, see the model page , or query the Models API. The model ID prefix is z-ai/ and can be omitted.

GLM models are served by Zhipu’s international site (zai); some models are also available from other providers such as Alibaba Cloud Model Studio and Volcano Engine, selected automatically by default. For how to select one, see Provider routing.

Examples

import os
from openai import OpenAI
 
client = OpenAI(base_url="https://api.ofox.io/v1", api_key=os.environ["OFOX_API_KEY"])
 
resp = client.chat.completions.create(
    model="z-ai/glm-5.1",
    messages=[{"role": "user", "content": "Explain idempotency in one sentence."}],
    extra_body={"thinking": {"type": "disabled"}},
)
print(resp.choices[0].message.content)

curl https://api.ofox.io/v1/chat/completions \
  -H "Authorization: Bearer $OFOX_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "z-ai/glm-5.1",
    "messages": [{"role": "user", "content": "Explain idempotency in one sentence."}],
    "thinking": {"type": "disabled"}
  }'

Thinking mode

All GLM models think by default. Thinking volume and latency we measured on the same math problem:

Passing "thinking": {"type": "disabled"} turns thinking off. In our tests it works on GLM-5.2, GLM-5.1, GLM-5, GLM-5 Turbo, GLM-5V Turbo and GLM-4.7 FlashX, and cuts latency substantially (GLM-5, for example, went from about 4.5 minutes to about 10 seconds). With the OpenAI SDK, put it in extra_body.

Thinking tokens are billed at the output rate; the count is in usage.completion_tokens_details.reasoning_tokens.

Reading the thinking content

• In streaming responses, thinking content is in delta.reasoning_content, the same as Zhipu’s official API.
• In non-streaming responses, thinking content is in message.reasoning_details; concatenate the text of each item in order.

Use in Claude Code

GLM models support the Anthropic-native API at https://api.ofox.io/anthropic:

curl https://api.ofox.io/anthropic/v1/messages \
  -H "x-api-key: $OFOX_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "z-ai/glm-5.3",
    "max_tokens": 8000,
    "messages": [{"role": "user", "content": "Explain idempotency in one sentence."}]
  }'

• On the Anthropic API, thinking content is returned as thinking blocks, and "thinking": {"type": "disabled"} also turns thinking off (on the same models as above).
• When thinking is on, set a large enough max_tokens and set the client timeout to 600 seconds or more.

For full Claude Code setup steps, see Claude Code setup.

Images and PDF

Models that support image input use the standard image_url content block; GLM-5V Turbo also accepts PDFs via a file content block:

import base64
import os
from openai import OpenAI
 
client = OpenAI(base_url="https://api.ofox.io/v1", api_key=os.environ["OFOX_API_KEY"])
img_b64 = base64.b64encode(open("photo.png", "rb").read()).decode()
 
resp = client.chat.completions.create(
    model="z-ai/glm-5v-turbo",
    messages=[{"role": "user", "content": [
        {"type": "image_url", "image_url": {"url": f"data:image/png;base64,{img_b64}"}},
        {"type": "text", "text": "Describe this image."},
    ]}],
    extra_body={"thinking": {"type": "disabled"}},
)
print(resp.choices[0].message.content)

GLM-5.3, GLM-5.2, GLM-5.1, GLM-5 and GLM-5 Turbo are text-only; do not send them images.

JSON output and tool calling

For response_format, use {"type": "json_object"} and state “return JSON” and the required fields in the prompt; tool calling uses the standard tools parameter. For the syntax, see Structured output and Function calling.

Prompt caching

GLM automatically caches repeated request prefixes, with no changes to your request. The cached amount is in usage.prompt_tokens_details.cached_tokens, and the cached portion is billed at the cache read rate.

Common errors

For other general errors, see Error Handling.

Official docs

• Zhipu Open Platform · Docs 
• Z.ai · Developer docs 

The official docs describe behavior when calling Zhipu directly. When calling through OfoxAI, model IDs and endpoints on this page take precedence.

FAQ