openai_base_url config and OPENAI_BASE_URL env var are ignored — always calls api.openai.com

Bug

Codex CLI v0.118.0 ignores the openai_base_url config.toml setting and the OPENAI_BASE_URL environment variable. All requests go to https://api.openai.com/v1/responses regardless of configuration.

Steps to Reproduce

1. Create ~/.codex/config.toml:

model = "gpt-5.4-mini"

[model_provider_config]
openai_base_url = "http://localhost:4000/v1"

1. Run:

OPENAI_API_KEY=test codex exec -m gpt-5.4-mini --sandbox danger-full-access "Say hello"

1. Observe: requests go to api.openai.com, not localhost:4000. A local proxy on port 4000 receives zero requests.

Also tried: OPENAI_BASE_URL=http://localhost:4000/v1 env var (prints deprecation warning but still ignores it).

Also tried: -c 'model_provider_config.openai_base_url="http://localhost:4000/v1"' flag.

Expected

Requests should go to the configured base URL, enabling Azure OpenAI and other OpenAI-compatible endpoints.

Environment

• Codex CLI: 0.118.0
• macOS 15 (Apple Silicon)
• Node.js 25.6.1

Use Case

We're trying to use Azure OpenAI ($100K sponsorship credits) with Codex CLI as part of a Paperclip agent orchestration setup. Azure supports the Responses API (/openai/responses?api-version=2025-03-01-preview) and works fine via curl, but Codex CLI can't be pointed at it.