Reviews · OCTOBER 27, 2026
GPT-6.1 Terra ships at $1/$5, matching Astra on agents at a tenth the token price
OpenAI's new mid-tier model lands October 26, 2026 with computer-use skills, a 50.3% Terminal-Bench 4.0 score, and prices half those of GPT-5.6 Terra and GPT-6.1 Sol.
OpenAI shipped GPT-6.1 Terra on October 26, 2026 at $1 per million input tokens and $5 per million output, half the rate of GPT-5.6 Terra and GPT-6.1 Sol, and roughly one-tenth the token price of GPT-6 Astra on comparable agent work. Cached input lands at $0.05. The model (ID gpt-6.1-terra) ships inside ChatGPT Plus, Pro, Enterprise, Edu, and Codex, and it can read screens, click through apps, fill forms, and operate software without a human cursor driving it.
That last capability is the news. Mid-tier models have been getting cheaper for two years; what hasn't been cheap is computer use. OpenAI's own launch example hands Terra a stream of incoming invoices to type into accounting software, per The Next Web. Translate that into a small-team economic reality: the owner-operator work that drains a 5-to-30-person company (CRM data entry, proposal drafting, form completion, routine outreach) was previously flagship-tier spending. Now it isn't.
The benchmark picture is unusually clean. On OpenAI's Agents' Last Exam v1 at medium effort, Terra matches GPT-6 Astra's score (57.6%) at $0.29 per task versus Astra's $4.82, the ratio behind the "one-tenth" claim, as Kingy.ai documented from OpenAI's charts. On Terminal-Bench 4.0, Terra hits 50.3% at max effort, more than double GPT-5.6 Terra's 21.5%. Factual-error rate on OpenAI's difficult-prompt test at low effort falls from 53.6% to 16.7%.
Artificial Analysis, running its own evals, scores Terra 48 on Intelligence Index v4.3.2, four points below Sol max and tied with Sol medium. The firm also clocks a 50-point drop in AA-Omniscience hallucination rate (88% to 38%) with a 2-point accuracy gain. On MMMU-Pro, Terra max hits 85.5%, within a point of Sol and Astra.
The buried detail is Sol. At medium effort, Sol scores 52.7% on Terminal-Bench 4.0 at about $1.02 per task; Terra max costs $1.84 for 50.3%. On the Intelligence Index, Sol medium runs $0.21 per index task to Terra max's $0.40. OpenAI still recommends Sol or Astra for complex coding, and the structural read matches our earlier coverage of Sol's pricing position: the mid-tier isn't one model, it's a routing problem.
Context on how Terra exists at all: OpenAI cancelled GPT-6.1 Astra last month after internal safety test failures, per The Next Web, and classifies Terra as Critical for cybersecurity capability under its Preparedness Framework, carrying the GPT-6 Astra safeguards stack. The verdict from Kingy is the right one: "Terra deserves a serious trial for visual office work and routine agents. For difficult software engineering... the sensible next step is a matched evaluation."
The frontier didn't move. The floor did.
Sources
- https://openai.com/news/product-releases/
- https://artificialanalysis.ai/articles/gpt-6-1-terra-pushes-the-cost-efficiency-frontier
- https://thenextweb.com/news/openai-gpt-6-1-terra-computer-use
- https://www.bleepingcomputer.com/news/artificial-intelligence/openai-launches-gpt-61-terra-with-3x-faster-ai-agents-50-percent-lower-pricing/
- https://kingy.ai/blog/gpt-6-1-terra/