Wink Pings

OpenAI Releases GPT-6.1 Terra: One-Tenth the Price, 3x the Speed, but This Time the Focus Is Not Writing Code

OpenAI has released GPT-6.1 Terra, with standard API pricing only one-tenth that of GPT-6 Astra and task completion more than 3x faster. It is better at operating computers, understanding screens, and filling out forms, and is seen as an important step in the direction of personal agents.

OpenAI has released GPT-6.1 Terra.

The name continues the celestial series. Sol is the sun, Terra is the earth. The official positioning is clear: understand what is on the screen, perform actions across apps, and write natural text. The tagline is "From visual understanding to action."

Key scenarios include technical drawings, drafting proposals, filling out forms, and editing videos. Note that there are no coding keywords at all.

The numbers that really matter are price and speed. Standard API pricing is $1 per million input tokens and $5 per million output tokens. Compared with the flagship model GPT-6 Astra, the cost per token is one-tenth, and average task speed is more than 3x faster. On the Agents' Last Exam benchmark, Terra at medium reasoning effort is 15 percentage points higher than GPT-5.6 Terra, while costing only one-third as much; it is on par with Astra, but at one-tenth Astra's price.

Agents' Last Exam benchmark

OpenAI also cited factuality improvements: at low reasoning effort, the error rate dropped from 53.6% to 16.7%; at maximum reasoning effort it fell to 4.3%. Coding ability has not been neglected either: it scored 50.3% on Terminal-Bench 4.0, more than double the previous generation.

But what is more interesting is its positioning. AI practitioner Alex Finn noted on Twitter that OpenAI's marketing this time emphasized "filling out forms, editing videos, and drafting proposals." He believes the model is clearly designed for personal agents. He plans to connect Terra to his own agent Hermes, but will still use Astra for coding tasks.

He also had a line worth pondering: "The model best at coding wins enterprises; the model best at using computers wins humans." He is more optimistic about computer-using models than code-writing models.

Someone in the comments questioned: "Filling out forms? Let it get through a Spanish government website first." Another person said: "You should run evaluations before replacing anything; the new model fabricated delivery dates 6 times out of 5 runs in my sales agent." This user is clearly being sarcastic, but the attitude is real: low prices and high scores won't convince everyone.

Terra is already available to all Plus, Pro, Enterprise, and Edu users of ChatGPT Work and Codex. Developers can use gpt-6.1-terra via the API.

Terra's release may mark the start of OpenAI seriously betting on personal agents. Personal agents operate your computer for you, rather than just chatting with you. The coding race continues, but another track has already started.

发布时间: 2026-10-27 08:50