- Sources: OpenRouter model page, HN discussion
- Summary: The OpenRouter model page shows the OpenAI provider row marked 50 percent off, with input falling from 5.00 to 2.50 dollars per million tokens, output from 30.00 to 15.00 dollars per million, and cache reads from 0.50 to 0.25 dollars per million. The provider table carries five rows. The plain Azure row is at the undiscounted 5.00, 30.00, and 0.50 dollars per million, and the Azure EU, Azure US, and Amazon Bedrock US rows are above that at 5.50, 33.00, and 0.55 dollars per million. The model carries a 1,050,000 token context and was released on 2026-07-09, and OpenRouter lists Codex, Hermes Agent, pi, Claude Code, and OpenClaw as its largest consumers by token volume. This is marked developing because the discount was verified only from the marketplace listing, no OpenAI announcement of a price change was located, and marketplace promotions can be time-boxed.
- Why it matters: Agent tooling billed per token through the OpenAI route pays half the listed rate, while every other listed route for the same model is at or above list price.
- Follow-up: Check for an OpenAI price announcement and whether the discount expires.
send feedback on this story