OpenAI Launches GPT-6.1 Terra with 9x Lower Cost Than Competitors
OpenAI has released GPT-6.1 Terra, a new model optimized for low-cost complex computer operations and visual reasoning. It outperforms predecessors in Agents’ Last Exam and OSWorld 2.1 benchmarks, offering 3x faster execution and costs approximately one-ninth of Claude Opus 5.5 per task. Additionally, it significantly reduces factual error rates, making it ideal for enterprise automation.
Full Report
On October 27, OpenAI announced the launch of its new model, GPT-6.1 Terra, early in the morning. The model is designed to perform complex computer operation tasks at a lower cost, with particular strengths in visual reasoning and desktop software manipulation. Compared to GPT-5.6 Terra, the new model demonstrates significant improvements in factual accuracy and coding capabilities.
Terra primarily targets enterprises seeking to reduce automation costs, enabling them to handle tasks such as invoice processing and technical drawing interpretation. According to OpenAI, Terra can complete difficult professional tasks at a fraction of the cost associated with premium models.
On the Agents’ Last Exam benchmark, which evaluates performance on complex professional work, Terra achieved markedly higher scores. Test cases included constructing 3D models from architectural blueprints and cataloging film shots for restoration teams.
At medium reasoning intensity, Terra scored 15 percentage points higher than GPT-5.6 Terra while completing tasks at approximately one-third of the cost. When compared to GPT-6 Astra, Terra delivered comparable benchmark results but at roughly one-tenth of the price per token.
Terra also holds an advantage in speed. OpenAI states that its task execution speed is more than three times that of GPT-6 Astra and GPT-6.1 Sol. At maximum reasoning intensity, the cost per task is approximately one-ninth that of Claude Opus 5.5.
Desktop software automation represents another area where Terra has made progress. OSWorld 2.1 tests a model’s ability to use desktop applications through multi-step operational tasks. At medium reasoning intensity, Terra not only surpassed the previous best score of GPT-5.6 Terra but also did so at significantly lower cost per task.
In the MMMU-Pro visual reasoning benchmark, Terra outperformed GPT-5.6 Terra across all reasoning intensities. Despite its lower cost, its performance remained close to that of GPT-6 Astra and GPT-6.1 Sol.
OpenAI has also focused on reducing factual errors in the model to minimize the impact of incorrect responses on automated workflows. The company utilized a suite of internal benchmarks with specially designed prompts to identify weaknesses in the model’s answers.
Results showed that at low reasoning intensity, Terra’s error rate was 16.7%, significantly lower than GPT-5.6 Terra’s 53.6%—a reduction of 36.9 percentage points.
When reasoning intensity was increased to the highest level, Terra’s error rate dropped further to 4.3%, 18 percentage points lower than the previous generation, approaching the levels of GPT-6 Astra and GPT-6.1 Sol.
Coding capabilities have also seen notable improvements. Terminal-Bench 4.0 tests AI coding agents’ problem-solving abilities via command-line tasks, such as fixing layout shifts during web page loading.
At maximum reasoning intensity, Terra successfully completed 50.3% of the test tasks, whereas GPT-5.6 Terra had a success rate of only 21.5%. This means the new model’s success rate is more than double that of its predecessor. For the most difficult coding tasks, OpenAI still recommends using GPT-6.1 Sol or GPT-6 Astra.
GPT-6.1 Terra became available to Plus, Pro, Enterprise, and Edu users on October 27, accessible within ChatGPT Work and Codex. Developers can also call the model via the OpenAI API using the identifier gpt-6.1-terra.
Regarding standard API pricing, it costs $1 per million input tokens (approximately ¥6.7 CNY at current exchange rates), $0.05 per million cached input tokens (approximately ¥0.33 CNY), and $5 per million output tokens (approximately ¥33.4 CNY). OpenAI notes that the new model is priced 50% lower than both GPT-5.6 Terra and GPT-6.1 Sol.
On the security front, Terra inherits the safeguards from GPT-6 Astra. OpenAI reports enhanced resistance to prompt injection attacks and improved clarity in communicating its own limitations to users. Under OpenAI’s preparedness framework assessment, Terra’s cybersecurity capabilities have reached the "Critical" threshold.
The lower price point is expected to make complex computer automation technologies more accessible to developers and enterprises. However, for the most challenging coding tasks, GPT-6.1 Sol and GPT-6 Astra remain the superior choices due to their stronger processing capabilities.