GPT-6.1 Terra is OpenAI's cheaper bet on vision and computer use

OpenAI released GPT-6.1 Terra, which it describes as a fast, efficient model built for detailed visual reasoning and precise computer use, with clearer writing than the model it replaces. The pitch is specific: reading technical drawings, filling out forms, drafting proposals, and video editing. Pricing is $1 per million input tokens and $5 per million output tokens, roughly half the cost of GPT-5.6 Terra and GPT-6.1 Sol, and about ten times the price of the cheapest GPT-6 Luna.

Simon Willison walked through where it fits. He compares Terra's position to Sonnet in Anthropic's lineup, neither the cheapest nor the smartest, which makes the choice harder rather than easier. His argument is that computer use is a good match for a cheaper model, because it tends to run as automated, repetitive work where speed and price matter more than peak intelligence. OpenAI's Ted Sanders, writing on Hacker News, framed Terra as a midrange API option: probably better than Astra and Sol at technical communication, similar on vision and computer use but much faster, and somewhat weaker elsewhere. For subscribers he called it "not a bombshell release."

Willison ran his usual pelican-on-a-bicycle test across reasoning levels. The max-reasoning result was a clean image with the frame, feet, and pedals in the right places, at a cost of 11.3 cents and 4 minutes 11 seconds, which also shows the latency that comes with turning reasoning up.

Why it matters

If you pick API models for a living, Terra is the new default to test for vision and computer-use pipelines that run at volume: half the price of Sol, much faster, and good enough on the tasks it targets. The tradeoff is real, so benchmark it against Sol on your own workload before you move spend over, especially where output quality on harder reasoning actually matters.