Running Codex on Unlimited Claude

Codex runs long autonomous loops, which is exactly the workload where per-token billing gets scary. Point it at a flat-rate unlimited Claude plan and let Codex iterate as long as a task needs — the price is the same whether it makes ten calls or ten thousand.

Set up the custom provider

Codex reads ~/.codex/config.toml. Add a [model_providers.<name>] table with the gateway base_url, the env_key holding your key, and wire_api = "chat" for the OpenAI-compatible surface, then select it with model_provider.

The CLI, the SDK, and the VS Code extension all read the same file, so one config covers every way you run Codex.

Base URL and model

Use the base URL with /v1 — Codex appends the route itself. Set model to a Claude id the gateway serves (claude-sonnet-4-5 or claude-opus-4-8), no prefix, and export your key into the env_key variable.

Provider ids openai, ollama and lmstudio are reserved; give yours a different name. Keys live in the user-level config, not a project copy.

Why unlimited fits Codex

Each Codex task is a multi-turn loop that resends context every step, so a single deep run can move enormous token volume. On metered pricing that is unpredictable — a loop that retries overnight can quietly run up a large bill.

Flat-rate unlimited removes the meter entirely. You stop rationing turns and let Codex pursue the task to completion, knowing the cost is fixed for the period.

Pick the right model for the loop

Use a Sonnet-class model for most agentic work and an Opus-class one for the hardest reasoning. On an unlimited plan you can leave the stronger model selected without watching token cost.

Verify wire_api against your Codex version — "chat" for OpenAI-compatible; some builds expect "responses".

# ~/.codex/config.toml
model = "claude-sonnet-4-5"
model_provider = "aiprime"

[model_providers.aiprime]
name = "AI Prime Gateway"
base_url = "https://claudeapikey.dev/v1"
env_key = "AIPRIME_API_KEY"
wire_api = "chat"
# export AIPRIME_API_KEY=<your key>

Frequently asked questions

Can I run Codex on an unlimited Claude plan?
Yes — define a custom provider in config.toml with the gateway base_url and wire_api = "chat", set a Claude model, and the flat-rate plan covers every run.

Why is unlimited good for Codex?
Codex runs long multi-turn loops that resend context, so token use is high and unpredictable. Unlimited keeps cost fixed regardless of run length.

Does the base URL need /v1?
Yes — {SITE}/v1. Codex appends the route itself.

Are there limits on the unlimited plan?
No token limits during the plan, only fair-use rate limits to keep the gateway fast.

AI Prime Tech is an independent API gateway. It is not affiliated with, endorsed by, or a reseller of Anthropic. Claude and related model names are trademarks of their respective owners.