- Published on
Just In: GPT-6.1 Terra Launches at a Rock-Bottom Price, Bringing the Worker Model for the Agent Era
- Authors
- Name
- DeepHorus
OpenAI has just officially released its next-generation model GPT-6.1 Terra, which focuses on faster responses, stronger visual understanding and computer operation capabilities, and lower usage costs. Compared with the previous generation GPT-5.6 Terra, the new model's standard API price is directly reduced by 50%. OpenAI positions it as an efficient model for high-frequency work scenarios, aiming to enable more enterprises and developers to apply AI Agents to everyday work at lower cost.
The price adjustment is the most direct signal of this release. GPT-6.1 Terra is priced at $1 per million tokens for input, $0.05 for cached input, and $5 for output. Compared with the flagship model GPT-6 Astra, its per-token unit prices for both input and output are only one-tenth of the latter's, yet in some professional task tests it has already come close to Astra's performance.
In terms of performance, GPT-6.1 Terra further combines visual understanding with computer operation, and can handle tasks such as reading technical drawings, filling out forms, editing videos, and operating desktop software. On the Agents' Last Exam test, at medium reasoning effort its score is 15 percentage points higher than GPT-5.6 Terra, while the cost per task drops to about one-third. Its task completion speed is more than 3 times that of GPT-6 Astra and GPT-6.1 Sol, and developer tests show that generation speed can reach 450 tokens per second in a single request.
Factual accuracy is another notable improvement. OpenAI's internal high-difficulty factual accuracy test shows that GPT-5.6 Terra's error rate at low reasoning effort was as high as 53.6%, while GPT-6.1 Terra has dropped to 16.7%, a decrease of 36.9 percentage points; at the highest reasoning effort, the error rate further drops to 4.3%. For Agents that handle contracts, technical documents, and enterprise internal data, this change is directly related to whether users can confidently hand over more work to them.
Coding ability has likewise improved substantially. On Terminal-Bench 4.0, GPT-6.1 Terra achieves a task resolution rate of 50.3% at the highest reasoning effort, while the previous generation's published result was only 21.5%, more than double. On Terminal-Bench-Science 0.1, it even surpasses the higher-positioned GPT-6.1 Sol, trailing Claude Opus 5.5 by less than 5 percentage points, while its estimated cost is less than one-fifth of the latter's.
On the product side, GPT-6.1 Terra is now available to ChatGPT Plus, Pro, Enterprise, and Edu users, and can be used in ChatGPT Work and Codex. Developers can also call it through the OpenAI API under the model name gpt-6.1-terra. OpenAI says that with the same usage quota, users can complete more tasks with the new model than with GPT-5.6 Terra.
In terms of safety, GPT-6.1 Terra reaches the Critical risk level for cybersecurity capability in OpenAI's Preparedness Framework, so it adopts the relevant safeguards of models such as GPT-6 Astra and inherits the alignment improvements of the GPT-6 series. From Astra to Sol to Terra, OpenAI's product division of labor is becoming clearer. As Agents enter real work scenarios, the focus of competition is extending from benchmark scores to task cost, speed, and reliable execution. Cheaper models with near-flagship capabilities are providing the conditions for AI Agents to move from a small number of high-value tasks toward large-scale daily applications.
