GPT-6.1 Terra vs GPT-6 Luna

The efficient end of the four models OpenAI recommends for Work and Codex. OpenAI points Luna at focused, high-volume tasks with a clear right answer — extraction, classification, structured summaries — and Terra at writing, visual understanding and computer use. The table shows what each choice costs.

Price and throughput favor Luna

Luna is far cheaper on every price cell, at every tier both publish. It also sits in the higher-throughput Luna rate-limit group: larger token ceilings than Terra's at every paid tier, the same requests per minute at Build and Launch, and more at Grow. For volume, Luna is the cheaper and the faster-to-scale option.

Caching cuts both ways

Terra bills a cached read at 0.05x its input rate, a deeper discount than Luna's 0.1x. Luna's input is so much cheaper, though, that its cached reads still cost less in absolute terms — the discount narrows Terra's gap on cache-heavy work without closing it.

What matches

The context window and maximum output, the long-context rule, the effort range from none to max with medium as the default, and Fast mode with EU data residency. Neither has Ultrafast. Tool calling splits slightly: Terra needs the Responses API for any tool call, while Luna also allows function calling in Chat Completions with effort set to none.

What favors Terra

A newer knowledge cutoff, and the kind of work OpenAI built it for: screens, images and prose, where a cheaper model's misses cost more than its tokens save.

Inside Codex

Luna is the Codex model on Free and Go; Terra starts at Plus and needs CLI 0.159.0 or later. OpenAI's five-hour estimates give Luna by far the larger message range on Plus, and its credit rate is the lowest of the four recommended models.

How to decide

Start well-specified, repeatable work on Luna. Move a task to Terra when Luna's output on visual or writing-heavy work isn't good enough — and judge that on finished tasks, not per-token price.