A dollar per million input tokens. At that price OpenAI ships a model that, in its own words, does computer use precisely.
Fact
The facts: on October 26, 2026, OpenAI's API changelog records the release of GPT-6.1 Terra - in the company's words a fast, efficient model for visual reasoning, precise computer use, and clear, natural writing. OpenAI's official feed announces it with the description Bringing fast and capable computer use to everyone. The model page lists prices per million tokens: input $1, cached input $0.05, output $5. The context is 1,050,000 tokens, with input up to 922,000 and output up to 128,000. The model takes text and images, returns text only, and has no audio or video. The identifier is gpt-6.1-terra, through Responses and Chat Completions. According to Vercel, which added it to its AI Gateway the same day, Terra improves on the previous Terra in visual reasoning, factual accuracy, and coding; the gateway also lists a faster variant, openai/gpt-6.1-terra-fast. Vercel adds that OpenAI still recommends its larger GPT-6 models for the most complex coding tasks.
The most honest sentence in the whole story comes through Vercel, and it is about the model's limit: for the most complex code, OpenAI still recommends the larger GPT-6 models. To Terra, Vercel leaves the bulk work with documents and desktop apps: forms, invoices.
A dollar for input, but not for the hardest code.
I could not open OpenAI's announcement for the release: the server refuses it. That is why there are no benchmarks. I also do not know whether Terra is in ChatGPT, or for whom.
Run the same screen test on it ten times and count the mistakes. The price will not tell you that.
Official primary sources
OpenAI - API changelog, entry of 26.10.2026OpenAI - GPT-6.1 Terra model pageOpenAI - official feed, Introducing GPT-6.1 Terra, 26.10.2026Vercel - GPT-6.1 Terra on AI Gateway, 26.10.2026The visual is generated code-art, no third-party images.