Configure LongCat-2.5 in Codex

One-sentence takeaway

LongCat officially provides a configuration for using LongCat-2.5 in Codex through the Responses API. The example uses the high reasoning level.

Use cases

• Suitable tasks: Coding tasks using Codex in the terminal with LongCat-2.5.

• Unsuitable tasks: Inferring the model's performance or cost on specific tasks from the configuration.

• Applicable model version: LongCat-2.5.

• Applicable client, agent, or API: Codex CLI; LongCat's OpenAI-format API.

• Recommended reasoning level and parameters: The official example sets model_reasoning_effort = "high", disables web search and response storage, and enables reasoning summary support.

Ready-to-use content

Edit ~/.codex/config.toml. The official example is:

model_provider = "codex"
model = "LongCat-2.5"
disable_response_storage = true
web_search = "disabled"
model_reasoning_effort = "high"
model_supports_reasoning_summaries = true

[model_providers.codex]
name = "codex"
base_url = "https://api.longcat.chat/openai/v1"
wire_api = "responses"
requires_openai_auth = true

Edit ~/.codex/auth.json and replace the placeholder with your own LongCat API key:

{
  "OPENAI_API_KEY": "<Your LongCat API Key>"
}

Test or workflow steps

1. Install Node.js 18.0 or later, run npm install -g @openai/codex, then check the installation with codex --version.

2. Get an API key from the LongCat platform and confirm on the usage page that you have sufficient quota.

3. Configure the model and authentication as shown above. Store the key securely; do not commit it to a version control repository.

4. Open a new terminal window and run codex. The official documentation uses the conversation interface opening normally as the startup check.

Raw evidence and data

• The model table in the source lists LongCat-2.5 and OpenAI / Anthropic as interface formats.

• The Codex configuration specifies wire_api = "responses" and LongCat's OpenAI-compatible base URL.

• The page provides no specific coding task, output, performance data, or Codex version number.

Scope and limitations

This is an official integration configuration, not a complete prompt or an independent performance test. The page does not describe the effects of other reasoning levels or verify compatibility across all Codex versions. Actual availability depends on the current client, account quota, and service status.

Reproduction notes

Install and configure as described in the source, then run codex in a new terminal. This article preserves the official configuration fields; the authentication value is a placeholder. The model was not called with a real key, and the conversation result was not verified during collection.