GPT-6.1 Terra brings lower API prices together with reported gains in several work-focused tasks. OpenAI positions it as a mid-range model for activities such as drafting proposals, interpreting technical drawings and filling out forms, as well as controlling software. Benchmark results described by Artificial Analysis suggest it can deliver stronger performance per dollar than its predecessor.
Lower prices, more capable performance
GPT-6.1 Terra’s API rates are half those of GPT-5.6 Terra: input tokens cost $1 per million and output tokens cost $5. Cached inputs cost $0.05 per million tokens, a quarter of the previous rate. For comparison, the article lists Claude Sonnet 5.5 at $2 per million input tokens and $10 per million output tokens.
Artificial Analysis reports that at maximum reasoning effort, Terra gains 6 points on the Intelligence Index over GPT-5.6 Terra. The estimated cost per task falls from $1.40 to $0.40. At lower effort levels, the reported performance gains are larger, at 7 to 13 points. Artificial Analysis describes the resulting balance of cost and performance as reaching the Pareto frontier.
That combination may matter for routine work where a model is used repeatedly. Lower prices can reduce the cost of processing many tasks, while improved results may limit the need to move each request to a larger model. The source’s figures are benchmark estimates, so they describe tested tasks rather than every real-world use.
Better results across coding and visual tasks
On the Coding Agent Index v2.0, GPT-6.1 Terra scores 48. That is below GPT-6.1 Sol’s 54, but above GPT-6 Luna’s 31. On Terminal-Bench 4.0, Terra solves 50.3 percent of tasks at maximum reasoning effort, compared with 21.5 percent for GPT-5.6 Terra. Artificial Analysis says the improvement puts it in the same tier as GPT-6.1 Sol on that benchmark.
Terra also performs well on scientific coding. On Terminal-Bench-Science 0.1, it outperforms GPT-6.1 Sol at lower cost and comes within five points of Claude Opus 5.5 for less than a fifth of the estimated cost. OpenAI still recommends GPT-6.1 Sol and GPT-6 Astra for the most demanding coding tasks, which marks a boundary to Terra’s intended role.
For visual reasoning, the model outperforms its predecessor at every reasoning level on MMMU-Pro and nearly matches GPT-6 Astra and GPT-6.1 Sol. On OSWorld 2.1, a desktop agent benchmark, Terra beats GPT-5.6 Terra’s best result at medium effort while costing a fraction as much.
Designed to operate software efficiently
OpenAI emphasizes automated software control as one of Terra’s strengths. On Agents’ Last Exam, the model matches GPT-6 Astra’s performance at one-tenth the token price. The source also reports that, compared with Claude Opus 5.5 at maximum effort, Terra is more than three times as token-efficient and uses a quarter of the token price, for roughly nine times lower cost per task.
OpenAI says Terra finishes more than three times faster than GPT-6 Astra and GPT-6.1 Sol. Faster completion can be useful when an agent needs to move through software interactions, but speed alone does not establish whether a model will handle every interface or task reliably. The reported results focus on benchmark comparisons.
OpenAI describes Terra as combining fast visual reasoning, precise computer control and a more natural writing style. The intended applications range from office tasks to interpreting technical drawings, tying its reasoning and computer-use capabilities to practical workflows.
Fewer unsupported guesses, with safeguards
A separate reported change concerns how often Terra guesses when it does not know an answer. On the private AA-Omniscience benchmark, it guesses 38 percent of the time in those cases, down from 88 percent for GPT-5.6 Terra. Accuracy also rose by two points. This suggests improved restraint on that test, though it does not establish how the model behaves in every situation.
The article says GPT-6.1 Terra meets OpenAI’s “Critical” cybersecurity threshold under the company’s Preparedness Framework and inherits GPT-6 Astra’s safeguards. According to the system card, Terra performs about as well as GPT-6 Astra and GPT-6.1 Sol against direct attacks on the instruction hierarchy; those models reach 99.99 percent on such tests.
For indirect prompt injection attacks, where malicious instructions are hidden in third-party content, Terra’s defense rate is 98.8 percent, up from 96.2 percent for GPT-5.6 Terra. OpenAI also says the model is more likely than its predecessor to stay within assigned boundaries and disclose restrictions to users.
GPT-6.1 Terra is available to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex, as well as through the API. Its pricing and benchmark results make it a candidate for everyday work and software control, while OpenAI’s own guidance reserves its largest models for the hardest coding tasks.