They don't actually seem to charge more for the >200k tokens on the API. OpenRouter and OpenAI's own API docs do not have anything about increased pricing for >200k context for GPT-5.4. I think the 2x limit usage for higher context is specific to using the model over a subscription in Codex.