How to use OpenAI Codex CLI with Unquant

Codex CLI talks to the OpenAI Responses API. Unquant serves a Responses-compatible endpoint, so Codex connects with a few lines of config and no proxy.

Setup guide2 min readUpdated 1 October 2026

1
Set your key
Codex reads the key from an environment variable you name in the config:
```
export UNQUANT_API_KEY=your-unquant-key
```

Add Unquant as a provider

Edit ~/.codex/config.toml:

model = "glm-5.3-flash"
model_provider = "unquant"

[model_providers.unquant]
name = "Unquant"
base_url = "https://api.unquant.io/v1"
env_key = "UNQUANT_API_KEY"
wire_api = "responses"

3
Run Codex
Run codex in your project. To use Qwen instead, set model = "qwen-3.8-flash-next".

Reasoning effort

Codex's model_reasoning_effort setting is passed through as reasoning effort: low or high on both models.

Troubleshooting

401

Unauthorized

The variable named in env_key must be set in the shell that runs Codex.

404

Not found

Use https://api.unquant.io/v1 as base_url and wire_api = "responses".

429

Concurrent stream limit reached

Another request is still running on your account. Let it finish, then retry.