Skip to content
Main Site News Console

Introducing GPT-6.1 Terra

· OpenAI Translated
OpenAILLM

Delivering Fast and Powerful Computer Use at Scale

We are officially introducing GPT-6.1 Terra, a fast, efficient model that combines fine-grained visual reasoning with precise computer use capabilities, alongside clearer and more natural text generation. Compared to GPT-5.6 Terra, this model represents a major leap forward, excelling at the heavy day-to-day communication and computational tasks that define modern workflows—from drafting proposal reports and interpreting technical drawings to automating form fills and video editing.

Terra continues to push the cost-intelligence frontier across our model families, striking a new balance between speed, performance, and cost. With standard API pricing reduced by 50% compared to GPT-5.6 Terra and GPT-6.1 Sol, developers can automate high-volume workflows with viable economics—such as parsing incoming invoices and entering line items directly into accounting systems. In ChatGPT Work and Codex, GPT-6.1 Terra also helps users get far more out of their plan allowances—supporting more tasks per usage quota compared to GPT-5.6 Terra, and significantly more than GPT-6.1 Sol or GPT-6 Astra.

From Visual Understanding to Action

We evaluated GPT-6.1 Terra’s computer use capabilities on Agents’ Last Exam, a professional task benchmark spanning complex workflows such as converting architectural blueprints into 3D models and reviewing animated films frame-by-frame to generate timecoded logs for restoration teams. At medium reasoning effort, GPT-6.1 Terra scored 15 percentage points higher than GPT-5.6 Terra at roughly one-third of the cost. Even compared to GPT-6 Astra, it delivers competitive performance at one-tenth of the per-token price. At maximum reasoning effort, Terra is over 3x more token-efficient than Claude Opus 5.5 at one-fourth the price per token, reducing overall per-task cost by approximately 9x. Terra is also significantly faster, completing tasks more than 3x faster than GPT-6 Astra and GPT-6.1 Sol.

On the visual reasoning benchmark MMMU-Pro, GPT-6.1 Terra outperforms GPT-5.6 Terra across all reasoning settings. Even with dramatically lower costs, its performance nearly matches our most capable frontier models, GPT-6 Astra and GPT-6.1 Sol.

OSWorld 2.1 evaluates an agent’s ability to interact with desktop applications to complete multi-step tasks (such as building 3D models from technical schematics). At medium reasoning effort, GPT-6.1 Terra surpasses GPT-5.6 Terra’s previous high score at a fraction of the cost per task.

Significant Improvements in Factual Accuracy and Coding

Over the past year, we have invested heavily in reducing model hallucinations and factual errors. On our internal benchmark designed with error-inducing prompts, the error rate at low reasoning effort dropped sharply from 53.6% on GPT-5.6 Terra to 16.7% on GPT-6.1 Terra—a reduction of nearly 37 percentage points. At maximum reasoning effort, the error rate dropped an additional 18 percentage points to 4.3%, nearly matching the performance of GPT-6 Astra and GPT-6.1 Sol.

In software engineering, GPT-6.1 Terra represents another major leap over GPT-5.6 Terra. Terminal-Bench 4.0 evaluates coding agents’ ability to resolve issues within command-line environments (such as fixing layout shift jitter during webpage loading). At maximum reasoning effort, GPT-6.1 Terra achieved a 50.3% task resolution rate—more than double GPT-5.6 Terra’s previously reported 21.5%. For exceptionally complex programming tasks, GPT-6.1 Sol and GPT-6 Astra remain the preferred choices.

Terminal-Bench-Science 0.1 evaluates an agent’s proficiency in using code to conduct scientific research tasks, such as tracking cell division in time-lapse microscopy videos. At maximum reasoning settings, GPT-6.1 Terra outperformed GPT-6.1 Sol at a lower cost, and narrowed the gap with Opus 5.5 to within 5 points at less than one-fifth of the estimated cost.

Safety and Alignment

As we work to make capable AI more efficient and accessible, we continue to train, test, and refine our safety guardrails. GPT-6.1 Terra meets the “Critical” threshold in our Preparedness Framework cybersecurity evaluations, carrying forward the protections established for GPT-6 Astra.

GPT-6.1 Terra also incorporates alignment improvements from the GPT-6 family. Evaluations show that compared to GPT-5.6 Terra, GPT-6.1 Terra adheres more strictly to task scope boundaries and transparently communicates its limitations and potential obstacles to users. In prompt injection evaluations, GPT-6.1 Terra demonstrates robust defense against malicious instructions designed to steer it away from the user’s designated task. For more details, see the system card.

Pricing and Availability

Starting today, GPT-6.1 Terra is available in ChatGPT Work and Codex to all Plus, Pro, Enterprise, and Edu users. Developers can also access the model via the OpenAI API under the model name gpt-6.1-terra.

Standard API pricing is: $1.00 per million input tokens, $0.05 per million cached input tokens, and $5.00 per million output tokens.