OpenAI puts Orion‑2 reasoning model in the API today
The model ships with a 400,000-token context window and, according to one reporter, sits below the company's previous frontier tier on price.
OpenAI released Orion-2, a reasoning model the company said is frontier-class, and made it available in its API on the day of announcement. OpenAI gives the context window as 400,000 tokens. The company did not mention a staged rollout or a waitlist.
Reporting from The Information's Stephanie Palazzolo adds a detail the company did not state directly: that Orion-2 is priced below OpenAI's previous frontier tier. Andrej Karpathy, writing separately, pointed to the reasoning benchmark results rather than the context window as the more consequential change, calling the evaluation jump the real story.
The two claims sit in different registers and should be read that way. The context window and availability come from OpenAI itself. The pricing position is a reporter's characterisation, and the benchmark assessment is one researcher's reading of results OpenAI published.