OpenAI Releases GPT-6.1 Terra With Lower Prices and Stronger Agent Benchmark Results
OpenAI has introduced GPT-6.1 Terra, a fast, efficient model that combines visual reasoning, computer use and natural writing in one system. The company describes it as a significant step up from GPT-5.6 Terra for communication and computer-based work. OpenAI
On pricing, OpenAI lists GPT-6.1 Terra's standard API rates (the per-token fees developers pay to build the model into their own software) at 50% below those of GPT-5.6 Terra and GPT-6.1 Sol. The company also says the model completes tasks more than three times faster than GPT-6 Astra and GPT-6.1 Sol.
The benchmark figures OpenAI reports are as follows. On Agents' Last Exam, a test of multi-step tasks carried out by AI agents, at medium reasoning effort (how long the model deliberates before answering), GPT-6.1 Terra scores 15 percentage points higher than GPT-5.6 Terra at about one-third of the cost. On MMMU-Pro, a test that mixes images and text, it outperforms GPT-5.6 Terra at every reasoning setting. On Terminal-Bench 4.0, which asks models to complete tasks in a command-line shell, it solves 50.3% of tasks at maximum reasoning effort, against 21.5% for GPT-5.6 Terra. On OpenAI's internal benchmark of error-inducing prompts (inputs designed to provoke mistakes) at low reasoning effort, the error rate fell from 53.6% with GPT-5.6 Terra to 16.7% with GPT-6.1 Terra.
OpenAI also reports that GPT-6.1 Terra meets the Critical threshold in cybersecurity under its Preparedness Framework, the company's internal system for rating risk. That categorisation is the reference point most relevant to teams deciding how to deploy the model in security-sensitive workflows.
Separately, OpenAI introduced GPT-6.1 Sol as an upgrade to GPT-6 Sol. The company says GPT-6.1 Sol nearly matches GPT-6 Astra's intelligence on agentic coding, computer use and professional work. On September 29, 2026, OpenAI posted an availability update naming GPT-6.1 Sol as its latest model. OpenAI
The lineup has moved quickly. In its earlier preview of the GPT-5.6 models, OpenAI described Terra as a balanced model for everyday work and listed it at $2.50 input and $15 output (input is the text sent to the model, output is the text it returns, and each is priced separately). It listed Luna at $1 input and $6 output. The company said Sol, Terra and Luna would become generally available in the coming weeks following the preview. OpenAI GPT-6 Sol and GPT-6 Luna are available in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise and Edu users.
In my view, the figures that matter most to engineers are the ones with a clear denominator. The Terminal-Bench move from 21.5% to 50.3% is large enough to change how teams think about agent loops that run shell tasks unattended. Still, a headline percentage on a single benchmark tells an engineer less than cost per completed task, latency under production load and the failure modes that remain.
The broader context here is the pricing. A model that is cheaper per token and faster to finish a task changes the economics of long-running agents more than it changes the leaderboard. Workloads that were too expensive to keep running, such as multi-step computer-use sessions and document-heavy review pipelines, become easier to justify at the same budget. The Critical cybersecurity rating points the other way, and it asks the same teams to put access controls, logging and monitoring in place before they widen deployment.
For now, this is a new model with aggressive pricing, strong reported gains on agentic and computer-use evaluations, and a cybersecurity classification that organisations will need to plan around. Production results will show how well those numbers hold.
