The /review command ignores the configured model and endpoint settings, defaulting to gpt-5-codex instead of the specified custom model. #5245

Environment

Codex Version: 0.46.0
OS: macOS 15.5 (24F74)
Shell: zsh 5.9
LiteLLM Version: 1.72.2

Configuration

~/.codex/config.toml:

model = "gpt-5_custom_model"
model_provider = "litellm"
approval_policy = "on-failure"
sandbox_mode = "danger-full-access"
tools.web_search = true
profile = "gpt5"

[model_providers.litellm]
name = "Chat Completions"
base_url = "https://api.example.com/v1"
env_key = "OPENAI_API_KEY"
wire_api = "chat"

[profiles.gpt5]
model = "gpt-5_custom_model"
model_provider = "litellm"
approval_policy = "on-failure"
request_max_retries = 4
stream_max_retries = 10
stream_idle_timeout_ms = 300000
model_reasoning_effort = "minimal"
model_reasoning_summary = "none"
model_verbosity = "low"
sandbox_mode = "danger-full-access"
tools.web_search = true

Environment Variables:

export OPENAI_API_KEY="$LITELLM_MASTER_KEY"
export OPENAI_API_BASE="https://api.example.com"
export OPENAI_BASE_URL="https://api.example.com"

Issue Description

The /review command ignores the configured model and endpoint settings, defaulting to gpt-5-codex instead of the specified custom model, despite the main interface correctly showing the configured model.

Expected Behavior

• /review should respect the model configuration from config.toml
• Should use the configured litellm provider with the custom base URL
• Should use the model specified in the active profile

Actual Behavior

Codex main interface correctly shows configured model:

╭─────────────────────────────────────────────────────────────────╮
│ >_ OpenAI Codex (v0.46.0)                                       │
│                                                                 │
│ model:     gpt-5_custom_model minimal   /model to change        │
│ directory: ~/Github/project-directory                           │
╰─────────────────────────────────────────────────────────────────╯

But /review command fails with wrong model:

>> Code review started: changes against 'main' <<
stream error: unexpected status 400 Bad Request: {"error":{"message":"{'error': '/chat/completions: Invalid model name passed in model=gpt-5-codex. Call `/v1/models` to view available models for your key.'}","type":"None","param":"None","code":"400"}}; retrying 1/5 in 191ms…
stream error: unexpected status 400 Bad Request: {"error":{"message":"{'error': '/chat/completions: Invalid model name passed in model=gpt-5-codex. Call `/v1/models` to view available models for your key.'}","type":"None","param":"None","code":"400"}}; retrying 2/5 in 376ms…
<< Code review finished >>

Steps to Reproduce

1. Configure Codex with custom model and LiteLLM provider as shown above
2. Verify main interface shows correct model configuration
3. Run /review command on any file
4. Observe that it attempts to use gpt-5-codex instead of configured model

Additional Context

• The main Codex interface correctly displays the configured custom model
• Other commands appear to respect the configuration correctly
• The issue is specifically with the /review command hardcoding gpt-5-codex
• Error occurs at the /chat/completions API endpoint level

Additional information

No response