How to use Codex with a cheaper API (OPENAI_BASE_URL guide)

OpenAI’s Codex hit 3 million weekly active users in 2026 — and for heavy users, the API cost is the main friction. Because Codex speaks the standard OpenAI API, you can route it through a cheaper, OpenAI-compatible gateway by changing one environment variable.

This guide shows how to use Codex with a cheaper API while keeping the real GPT models.

How it works

Codex sends requests to whatever OPENAI_BASE_URL points to. Point it at a unified gateway like nexinfer and your requests go to the same models at a lower price — with published multipliers and zero request logging.

Setup (2 steps)

1. Set the base URL and key

export OPENAI_BASE_URL=https://api.nexinfer.com/v1
export OPENAI_API_KEY=sk-nexinfer-xxxx   # your nexinfer key

Add them to your shell profile (~/.zshrc / ~/.bashrc) to persist.

2. Run Codex as usual

codex

Same workflow, lower cost. Streaming and tool calls are untouched.

Using the OpenAI SDK directly

If you call the API from code, only the base_url changes:

from openai import OpenAI

client = OpenAI(
    base_url="https://api.nexinfer.com/v1",
    api_key="sk-nexinfer-xxxx",
)
resp = client.chat.completions.create(
    model="gpt-5",
    messages=[{"role": "user", "content": "Hello"}],
)

What to check in a gateway

• Real models — gpt-5 means gpt-5, not a cheaper stand-in.
• Transparent billing — no hidden routing markup (some gateways quietly add 5%).
• No logging — your prompts and key stay yours.

Frequently asked questions

Will my existing OpenAI code work unchanged? Yes — it’s the standard OpenAI API. Only base_url and the key change.

Does it support GPT-5 and the mini models? Yes, plus Claude, Gemini and more through the same endpoint.

How much can I save? Up to ~35% versus official list price on GPT models — check live rates on the pricing section.