GPT-6.1 Terra (OpenAI API model page)

Summary

OpenAI released GPT-6.1 Terra (gpt-6.1-terra) on 2026-10-26. It is a mid-priced model aimed at visual reasoning, computer use, and clear writing. It sits between GPT-6.1 Sol and GPT-6 Luna on price. The announcement post ("Introducing GPT-6.1 Terra") returned 403 to this run, so the facts here come from the API model page, the API changelog of 2026-10-26, and Simon Willison's write-up.

Key points

Price per 1M tokens (2026-10-26): $1 input, $0.05 cached input, $5 output. A 30-minute cache write costs $1.25. That is half the price of GPT-6.1 Sol and about 10x GPT-6 Luna.
Limits: 1,050,000-token context, at most 922,000 input tokens, and 128,000 output tokens. Knowledge cutoff is 2026-06-01, later than other GPT-6 models.
Reasoning effort: none, low, medium (default), high, xhigh, max.
Tools in the Responses API: web search, file search, image generation, code interpreter, hosted shell, apply patch, skills, computer use, MCP, and tool search.
Endpoints: Responses, Chat Completions, and Batch. It is not available in the Realtime, Assistants, or fine-tuning APIs.
Quality, according to OpenAI: OpenAI's Ted Sanders (on Hacker News, quoted by Simon Willison) said Terra is better than Astra and Sol at technical communication, about as good at vision and computer use but much faster, and somewhat weaker at most other tasks. No benchmark figures were checked for this note.
Availability: in the API and in GitHub Copilot from 2026-10-26 (Pro, Pro+, Max, Business, Enterprise; billed at list price under usage-based billing, gradual rollout).

Why it matters

Terra is a cheaper choice for computer-use and vision agents that do repetitive work, where Astra or Sol cost too much and Luna is not capable enough. Its later knowledge cutoff also helps.