OpenAI releases GPT-6.1 Terra with 50% lower API costs

A new standard for office AI

OpenAI has officially released GPT-6.1 Terra, a new model specifically engineered for practical workplace applications. The company announced the launch on October 26, positioning the model as a faster and more affordable alternative to previous iterations. It is designed to handle a variety of everyday office tasks, including drafting proposals and interpreting technical drawings.

The model is built to excel in areas like visual reasoning, coding, and general computer use. This means it can interact with software and process visual data more effectively than standard text models. OpenAI intends for Terra to be the primary replacement for the older GPT-5.6 Terra model in many production environments.

Users can access the model through several existing platforms. It is available to Plus, Pro, Enterprise, and Edu subscribers within ChatGPT Work and Codex. For developers, the model can be accessed directly through the API using the specific identifier gpt-6.1-terra.

Drastic pricing and speed gains

One of the most significant aspects of the launch is the aggressive pricing structure. OpenAI has set the standard API prices for GPT-6.1 Terra to be half the cost of both GPT-5.6 Terra and GPT-6.1 Sol. Specifically, the model charges $1 for every million input tokens, $0.05 for every million cached input tokens, and $5 for every million output tokens.

The model also demonstrates substantial improvements in speed and efficiency compared to its predecessors. OpenAI reports that Terra completes tasks more than three times faster than both GPT-6 Astra and GPT-6.1 Sol. This speed makes it particularly well-suited for high-volume automation where time-to-completion is a critical factor for users.

When compared to competitors, the cost-to-performance ratio is notably high. At maximum effort, the company claims Terra is over three times as token efficient as Anthropic's Claude Opus 5.5. This efficiency translates to roughly 9x lower cost per task for users performing complex operations.

Benchmark performance and accuracy

OpenAI provided several internal benchmarks to demonstrate the model's capabilities. On the Agents' Last Exam, which tests professional computer-use tasks like turning architectural drawings into 3D models, GPT-6.1 Terra scored 15 percentage points higher than GPT-5.6 Terra. This improvement occurred at medium reasoning effort while costing about one-third as much.

The model also shows significant progress in reducing factual errors. At low reasoning effort, the error rate dropped from 53.6% with the previous Terra model to just 16.7% with GPT-6.1 Terra. At maximum effort, the error rate fell by 18 percentage points to 4.3%, which is nearly identical to the rates seen in the more expensive GPT-6 Astra and GPT-6.1 Sol models.

Coding capabilities also saw a measurable boost during testing. On the Terminal-Bench 4.0, which evaluates command-line coding agents, Terra solved 50.3% of tasks at maximum reasoning effort. This is a major jump from the 21.5% success rate recorded for GPT-5.6 Terra, though OpenAI still suggests using higher-tier models for the most complex coding needs.

Safety and practical applications

Beyond performance, OpenAI has integrated specific safety and security features into the new model. It meets the Critical threshold in cybersecurity under the company's Preparedness Framework. It also retains the protections originally developed for GPT-6 Astra and shows stronger resistance to prompt-injection attempts than the previous Terra version.

The practical applications for this model are centered on automating tedious manual work. OpenAI highlights tasks such as reading incoming invoices and automatically entering that data into accounting software. It is also designed to handle filling out forms, editing videos, and interpreting complex technical documents for various industries.

For existing subscribers, the launch means that the same plan allowance will cover more tasks than before. Because the model is more efficient, users of ChatGPT Work and Codex can perform more actions within their current limits. Developers are encouraged to switch their model identifiers to gpt-6.1-terra to compare the cost per task against their current production workloads.