How to Choose and Use a Codex Relay

// Why a Relay

1. Why Codex Needs a Relay

Connecting Codex CLI directly to OpenAI’s official API has three real obstacles in China: unstable network requiring a VPN, official support for overseas credit cards only, and higher cost at official pricing. A relay solves all three at once.

// Selection Criteria

2. Six Dimensions for Picking a Codex Relay

Do not judge by price alone. A cheap relay that rate-limits and throws 502s constantly costs more in practice than a slightly pricier stable one. Judge by the priority below.

1. Model Support: Does It Actually Support Codex / GPT-5

This is the first hard requirement. Some relays claim OpenAI support but only offer GPT-4o, not the models Codex CLI needs. Confirm it explicitly supports codex or GPT-5 and provides /v1/responses or a Codex-compatible endpoint.

2. Stability: Availability and Latency

Check whether it publishes availability (99.9% is the top standard) and has a status page. Poor stability shows as peak-hour 502s, rate limits, and slow responses. Prefer relays with multi-route failover.

3. Price Multiplier: Not Just the Number

Multiplier = relay price / official price. 1.0x is the official rate; below 1.0x, be wary of downgraded models. A reasonable range is 0.8–1.5x.

4. China Direct & Payment

Confirm it connects without a VPN and supports RMB payment. This is a must for developers in China.

5. Free Credit: Try Before You Pay

Legit relays usually give a small free credit. Always test that Codex works end-to-end with the free credit before topping up. Never make a large deposit upfront.

6. Invoicing & Compliance

For reimbursement, confirm it can issue invoices and has a stable company entity. Individual developers can skip this.

// Pitfalls

3. Common Pitfalls

• Fake Codex support: the page says "supports OpenAI" but Codex CLI cannot connect — always test with free credit first.
• Silent downgrade: claims GPT-5 but returns GPT-4o-mini quality; be especially wary of absurdly cheap relays.
• Deposit-and-vanish: new relays with no reputation pushing large deposits / bonus tiers. Test small for 1–2 weeks first.
• Opaque rate limits: no stated concurrency or rate limits, then frequent 429s in practice.
• No status page: when things break you cannot tell if it is your config or their outage.

// Setup

4. After Choosing: Codex CLI Setup

Once you pick a relay, connecting Codex CLI takes three steps: get the Base URL and API Key, set environment variables, and verify.

1. Install Codex CLI

npm install -g @openai/codex
# or
brew install codex

2. Configure Base URL and Key

Get the Base URL and API Key from the relay dashboard and set env vars:

export OPENAI_BASE_URL="https://your-relay-host/v1"
export OPENAI_API_KEY="sk-your-key"

Or write to Codex’s config file ~/.codex/config.toml:

[model_providers.relay]
name = "relay"
base_url = "https://your-relay-host/v1"
env_key = "OPENAI_API_KEY"

[profiles.default]
model_provider = "relay"
model = "gpt-5-codex"

3. Verify

codex "write a Python function for the Fibonacci sequence"

If it returns code, you are connected. A 401 means check the Key, a 404 means check the Base URL includes /v1, and a 429 means you are rate-limited.

// Decision Checklist

5. One-Line Decision Checklist

• Trying it out / budget-sensitive: pick a relay with free credit and 0.8–1.0x multiplier, get it working first
• Daily main development: stability first — pick a top relay with published 99.9% availability and a status page
• Enterprise / team: must issue invoices, have a company entity, and support team management
• Always: test Codex end-to-end on free credit, confirm no downgrade or rate limit, then top up

Choosing a relay is a balance of stability × price × compliance. Do not be lured by the lowest price — the relay that runs your tasks reliably long-term is the real bargain.