Codex
Superconductor is the best place to run Codex in the cloud.
Credentials
Superconductor supports two ways to authenticate Codex:
- ChatGPT plan: Connect your ChatGPT Pro or Plus account via OAuth in your Account, Project, or Workspace settings. Your connected plan is used when you launch an implementation directly. Implementations launched automatically or through integrations — such as Slack, GitHub, Email triage, the Meeting assistant, or external PR handling — use a configured API key instead.
- API Keys: Configure your OpenAI API key at the Project or Workspace level. Used for automatic and integration-triggered launches, and when a direct launch doesn't have a connected ChatGPT plan available.
See Agent credentials for details on precedence rules.
Models
Codex supports the following models:
- GPT-5.5 - OpenAI's newest coding model. See OpenAI's GPT-5.5 announcement for details. You can pick it explicitly on an implementation, or set it as your default for integrations.
- GPT-5.3 Codex Spark - Lightweight Codex variant optimized for faster execution. Requires a connected ChatGPT Pro plan.
Each model supports configurable reasoning effort levels: medium, high, and xhigh. This means there are six model variants available in total.
What are reasoning effort levels? Reasoning effort controls how much computation the model uses when thinking through a problem. Medium offers a good balance of speed and quality for most tasks. High increases deliberation time for better results on harder problems. Xhigh applies maximum reasoning effort, producing the best results for complex tasks but with higher cost and latency.
Learn more about OpenAI models in the OpenAI documentation.
Usage stats
If you've connected a ChatGPT Pro or Plus plan, Superconductor shows your rate limit utilization. See Usage stats for details.