OpenAI Launches GPT 6.1 Terra With Faster Computer Use, Lower Hallucination Rates, and 50 Percent Lower API Pricing
GPT-6.1 Terra: delivering fast and capable computer use at a lower price. GPT-6.1 Terra understands what’s on screen, takes action across apps, writes naturally, and is a more capable model for your whole workday with the efficiency to take on more of it.
Context
Delivering GPT 6 Astra level computer use and visual reasoning at one tenth the token cost and over three times the speed makes continuous desktop and browser agent automation economically viable for everyday enterprise workflows.
OpenAI introduced the Terra tier alongside its flagship Astra and Sol model families to serve high volume multimodal and computer use workloads at lower latency. Earlier iterations such as GPT 5.6 Terra struggled with complex command line engineering and higher factual error rates at low reasoning settings.
Deep dive
On October 26, OpenAI launched GPT 6.1 Terra, a fast multimodal model built for computer use, visual reasoning, and natural workplace writing across ChatGPT Work, Codex, and the OpenAI API. Priced at 1 dollar per million input tokens, 0.05 dollars per million cached input tokens, and 5 dollars per million output tokens, GPT 6.1 Terra cuts standard API pricing by 50 percent compared with both GPT 5.6 Terra and GPT 6.1 Sol.
On Agents' Last Exam, a benchmark measuring professional software workflows such as converting architectural drawings into 3D models and logging film restoration shots, GPT 6.1 Terra scores 15 percentage points higher than GPT 5.6 Terra at medium reasoning effort while costing roughly one third as much. OpenAI reports that Terra matches its frontier GPT 6 Astra model on the benchmark at one tenth the price per token, completes tasks over three times faster than both Astra and GPT 6.1 Sol, and achieves roughly nine times lower cost per task than Claude Opus 5.5 at maximum effort.
Beyond desktop navigation on OSWorld 2.1 and visual reasoning on MMMU Pro, GPT 6.1 Terra more than doubles its predecessor's command line coding score on Terminal Bench 4.0 from 21.5 percent to 50.3 percent. On OpenAI's internal hallucination benchmark, error rates dropped from 53.6 percent to 16.7 percent at low reasoning effort and down to 4.3 percent at maximum effort, while meeting the Critical cybersecurity threshold under OpenAI's Preparedness Framework.
What to expect next
Cutting computer use token costs to 1 dollar per million input tokens and 5 cent cached reads shifts GUI automation from expensive demos to high volume back office operations like invoice processing and legacy software entry. Competing labs will now face intense pressure to match Terra's three fold speed advantage and cached token economics while hardening desktop agents against prompt injection across untrusted application windows.