IT之家 News on October 27: OpenAI announced early this morning (the 27th) the launch of a new model, GPT-6.1 Terra, which focuses on completing complex computer operation tasks at lower cost, and is especially good at work requiring visual reasoning and operating desktop software. Compared with GPT-5.6 Terra, the new model also has major improvements in factual accuracy and programming ability.
Terra is mainly aimed at enterprises that want to reduce automation costs, and can be used to process invoices, interpret technical drawings, and other work. OpenAI said that when facing difficult professional tasks, Terra can complete them at a cost far lower than expensive models.
On the Agents’ Last Exam benchmark, which examines complex professional work, Terra's scores improved markedly. Test items include building three-dimensional models from architectural drawings, and logging film shots one by one for film restoration teams.
With medium reasoning effort, Terra scored 15 percentage points higher than GPT-5.6 Terra, while the cost of completing the tasks was only about one-third of the latter's. Compared with GPT-6 Astra, Terra's test results are on par, yet the price per Token is only about one-tenth of the latter's.
Terra also has an advantage in speed. OpenAI said thatits task execution speed is more than 3 times that of GPT-6 Astra and GPT-6.1 Sol. At the highest reasoning effort,the cost per task is also only about one-ninth that of Claude Opus 5.5。
Desktop software automation is another area where Terra has made progress. OSWorld 2.1 tests a model's ability to use desktop applications through multi-step operation tasks. With medium reasoning effort, Terra not only surpassed GPT-5.6 Terra's previous best result, but also completed each task at a noticeably lower cost.
On the visual reasoning benchmark MMMU-Pro, Terra's results at all reasoning effortsexceeded those of GPT-5.6 Terra. Despite the lower cost, its performance is still close to GPT-6 Astra and GPT-6.1 Sol.
OpenAI alsofocused on reducing the model's factual errorsto lessen the impact of wrong answers on automated processes. The company used an internal benchmark with specially designed prompts to probe weak points in the model's answers.
The results show that with low reasoning effort, Terra's error rate is 16.7%, far below GPT-5.6 Terra's 53.6%, a drop of 36.9 percentage points.
After raising the reasoning effort to the highest level, Terra's error rate fell further to 4.3%, 18 percentage points lower than the previous generation, approaching the level of GPT-6 Astra and GPT-6.1 Sol.
Programming ability has also improved significantly. Terminal-Bench 4.0 tests AI coding agents' problem-solving ability through command-line tasks, such as fixing layout shifts that occur when a web page loads.
At the highest reasoning effort, Terra successfully completed 50.3% of the test tasks, while GPT-5.6 Terra's success rate was only 21.5% — the new model's success rate is more than twice that of the previous generation. For the hardest programming tasks, OpenAI still recommends GPT-6.1 Sol or GPT-6 Astra.
GPT-6.1 Terra became available on October 27 to Plus, Pro, Enterprise, and Edu users, and can be used in ChatGPT Work and Codex. Developers can also call it through the OpenAI API, with the model identified as gpt-6.1-terra。
For standard API pricing, every 1 million input Tokens cost 1 US dollar (IT Home note: about 6.7 RMB at the current exchange rate), every 1 million cached input Tokens cost 0.05 US dollars (about 0.33 RMB), and every 1 million output Tokens cost 5 US dollars (about 33.4 RMB). OpenAI said the new model's price is 50% lower than GPT-5.6 Terra and GPT-6.1 Sol.
On safety, Terra adopts the safeguards of GPT-6 Astra. OpenAI said the new model has enhanced resistance to prompt injection attacks and can more clearly explain its limitations to users. Evaluated under OpenAI's Preparedness Framework, Terra's cybersecurity capabilityhas reached the "critical" threshold。
. The lower price is expected to make it easier for developers and enterprises to adopt complex computer automation technology. For the hardest programming tasks, GPT-6.1 Sol and GPT-6 Astra still have stronger handling capability.