# GPT-6.1 Terra arrives on launch day and leads Auto Fast

> OpenAI's GPT-6.1 Terra is a hosted agent model at $1 input and $5 output per million tokens, now first in Auto's Fast tier and offered on Token Tunnel.

Oct 26, 2026. Topics: Models. https://www.generalinput.com/changelog/2026-10-26-gpt-6-1-terra

OpenAI released GPT-6.1 Terra on October 26, and it's in General Input the same day. Every workspace can pick it for specialists, workflows, and chats, and it runs on our keys with no setup. Terra costs half of GPT-6.1 Sol per token, so it suits everyday agent work where Sol's extra depth isn't needed. It also becomes the first model Auto's Fast tier tries. If your ChatGPT plan includes Terra in Codex, [Token Tunnel](/token-tunnel) offers it on the plan you already pay for. Compare models on the [models page](/models).

### How it works

- **Pricing and window.** $1 input and $5 output per million tokens, and $0.05 per million cached input tokens, with 1.05M tokens of context and image and PDF input. A request over 272K input tokens is billed at OpenAI's long-context rate: twice the input price and 1.5 times the output price.
- **Effort.** Terra defaults to extra high, with none, low, medium, high, and max to pick from.
- **Auto Fast.** Fast now runs Terra at medium, then Claude Haiku 5.5, Gemini 3.8 Flash, and Grok 4.7. On Artificial Analysis's Intelligence Index, Terra at medium scores 41 to Haiku's 34 and starts answering in about 1.4 seconds. GPT-6 Luna leaves the tier. Balanced and Max are unchanged.
- **Routes.** OpenAI serves it first, with Vercel AI Gateway and OpenRouter behind it. When it's overloaded, runs fall back to Claude Sonnet 5.5, then Gemini 3.8 Flash.
- **Token Tunnel.** The tunnel lists Terra as soon as your signed-in ChatGPT account serves it, under the same name as the hosted model. Saved picks of the older GPT-5.6 Terra keep running on GPT-6.1 Sol.

### Also shipped

- Vercel AI Gateway now serves GPT-6.1 Sol and GPT-6 Luna as well, so each has a second backup route.