GLM-5.3

Item	Value
Base URL	`https://api.hop-base.com/v1`
Model	`glm-5.3`
Chat	`POST /v1/chat/completions` only — `/v1/responses` and `/v1/messages` are not supported and return 400

Use a key that includes the GLM Official Direct group; keys without it get a 404.

Group selection and billing

GLM-5.3 is Zhipu's latest flagship: 1M context window with thinking-length control, served via official direct access.

glm-5.3 is sold through the GLM Official Direct group, with a 1M context window. When creating or editing a key under API Keys in the console, add this group to the key. Your effective rate is shown in the signed-in model catalog.
Keys that include neither group get a clear 404 when calling glm-5.3 — add one of these groups to the key in the console, or create a new key with it.
GLM-5.2 has been discontinued; projects that called glm-5.2 only need to switch the model name to glm-5.3.

Request parameters

For POST /v1/chat/completions. The gateway only rejects empty messages; every other limit comes from the model service, with error text quoted from production records.

Parameter	Required	Type & limits	Default	Description
`model`	Required	`glm-5.3`	—	Exact match
`messages`	Required	Non-empty array, text only	—	Missing or empty returns a gateway 400; images return 400. Context is 1M
`max_tokens`	Optional	Integer 1–131,072	Service default	Outside the range returns 400 `max_tokens参数非法：限制数值范围[1,131072]`; shared by thinking and answer
`reasoning_effort`	Optional	`low` / `high` / `max`	Model default	Always thinks: `none` or any other thinking-off value returns 400
`temperature`, `top_p`	Optional	Passed through; no gateway range	Service default	Sampling controls, forwarded as-is
`tools`, `tool_choice`	Optional	As in OpenAI	`auto`	A returned `tool` result must match a call ID from the previous turn, or you get 400 `No tool call found for function call output`
`stream`	Optional	Boolean	`false`	`true` returns server-sent events
`stream_options.include_usage`	Optional	Boolean	`false`	Must be `true` to receive the `usage` chunk when streaming

The wording of the thinking-off 400 differs between groups, so handle it by status code and your reasoning_effort value, not the message text.

Client fields

Field	Value
API address / Base URL	`https://api.hop-base.com/v1`
API Key	`sk-your-key`
Model	`glm-5.3`
API format (CC Switch and other routing-capable clients only)	`OpenAI Chat Completions (routing required)`

The "API format" field above exists only in CC Switch and similar clients that can route between protocols — it is not a parameter you send in the request body. In Claude Code, CC Switch, and other clients that support routing, GLM-5.3 does not use the Anthropic protocol. In advanced settings, select OpenAI Chat Completions (routing required). A plain OpenAI SDK or curl call never touches this setting; skip straight to the sections below.

Use GLM-5.3 in Claude Code with CC Switch

Use CC Switch 3.17.0 or later. GLM-5.3 uses OpenAI Chat Completions, so CC Switch must translate requests through its local router into the Anthropic Messages protocol used by Claude Code.

Add a HopBase GLM provider

Open CC Switch, select the Claude Code tab, select + in the upper-right corner, then choose Custom configuration. Enter:

Provider name: HopBase GLM
API Key: your sk- key
Request URL: https://api.hop-base.com/v1

Set the API format and model mapping

Expand Advanced options, apply these settings, then save:

API format: OpenAI Chat Completions (routing required)
Authentication field: keep the default ANTHROPIC_AUTH_TOKEN
Default fallback model: glm-5.3

To force the Sonnet, Opus, and Haiku roles to use GLM-5.3, also map all three roles to glm-5.3. After saving, the provider card shows Routing required.

Enable local routing

Open Settings → Routing → Local routing. Enable the main routing switch, then enable Claude Code under application routing. Confirm the first-run message and leave CC Switch running in the background.

Enable and verify the provider

Return to the Claude Code provider list and select Enable on the HopBase GLM card. After enabling routing takeover for the first time, close and reopen the terminal, run claude, and send a short prompt.

Return to Settings → Routing. The setup works when Current Provider shows HopBase GLM and the request count increases. Once routing is active, later provider switches are hot-swapped without another terminal restart.

Codex CLI

glm-5.3 and glm-5.3-flash in the GLM Official Direct group only offer OpenAI Chat Completions: /v1/responses returns 400 endpoint_not_supported. Codex CLI has spoken only Responses since 0.122 (wire_api = "chat" was removed), so Codex CLI cannot connect to GLM-5.3. In Codex CLI, use GPT, DeepSeek, Qwen, or the US Region glm-5.3-us, which supports all three protocols — see US Region.

OpenAI SDK / Python

from openai import OpenAI

client = OpenAI(
    base_url="https://api.hop-base.com/v1",
    api_key="sk-your-key",
)

resp = client.chat.completions.create(
    model="glm-5.3",
    messages=[{
        "role": "user",
        "content": "Explain where GLM-5.3 works best",
    }],
)
print(resp.choices[0].message.content)

curl

curl https://api.hop-base.com/v1/chat/completions \
  -H "Authorization: Bearer sk-your-key" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "glm-5.3",
    "messages": [{"role": "user", "content": "Hello, introduce yourself"}]
  }'

Next steps

GLM-5.3 Flash — multimodal, lower unit price
CC Switch — installing and troubleshooting the local router
Error codes and retries
Concurrency, timeouts, and billing

On this page