OpenAI has unveiled GPT-6.1 Terra, a reasoning model that boasts near-frontier performance on professional tasks while offering a more affordable option. This new model replaces GPT-5.6 Terra and is geared towards organizations seeking to enhance accuracy and automation for large-scale tasks.
Addressing the needs of professional workloads
GPT-6.1 Terra occupies a middle ground between OpenAI’s more affordable GPT-6 Luna and its higher-end models, such as GPT-6.1 Sol and GPT-6 Astra. OpenAI emphasizes the model’s ability to utilize computers effectively, highlighting its capacity to draft proposals, interpret technical drawings, fill out forms, and edit videos.
Pareekh Jain, CEO of EIIRTrend & Pareekh Consulting, observes that GPT-6.1 Terra fills a specific gap in the market. “Enterprises need AI models that balance intelligence, speed, and cost,” Jain stated. “Frontier models are expensive for routine tasks, while smaller models may lack sufficient capabilities. Terra addresses this gap, competing with Anthropic’s Claude Sonnet and Google’s Gemini Flash models.”
Evaluating performance across various tasks
OpenAI substantiates its claims with benchmark scores in three key areas. On the Agents’ Last Exam benchmark, which assesses professional tasks, GPT-6.1 Terra achieved a score of 60.0% at a cost of $1.01 per task. This represents an improvement over GPT-5.6 Terra, which scored 50.7% at $3.58 per task. Although it trails the more expensive GPT-6.1 Sol, which scored 58.6% at $2.00 per task, and the flagship GPT-6 Astra, which scored 59.3% at $7.24 per task, GPT-6.1 Terra remains a competitive option. Claude Opus 5.5 led this category with a score of 63.2% at $8.82 per task.
The results of the MMMU-Pro benchmark, which evaluates visual reasoning, show GPT-6.1 Terra outperforming its predecessor. At maximum effort, GPT-6.1 Terra attained a score of 85.5% at $0.012 per task, compared to GPT-5.6 Terra’s score of 81.7% at $0.079. Within OpenAI’s current lineup, GPT-6 Luna scored 79.9% at $0.0021 per task, GPT-6.1 Sol achieved 86.3% at $0.013 per task, and the flagship GPT-6 Astra led with a score of 86.7% at $0.065 per task.
On the OSWorld 2.1 benchmark, which assesses how AI agents handle desktop applications, GPT-6.1 Terra scored 74.2% at maximum effort, costing $1.19 per task. This represents a significant upgrade over GPT-5.6 Terra’s score of 57.1% at $4.76 per task. However, it still trails Anthropic’s Claude Sonnet 5.5, which led the category with a score of 83.9% at $5.74 per task.
Examining the economics and pricing structure
GPT-6.1 Terra is priced at $1 per million input tokens, $5 per million output tokens, and $0.05 per million cached input tokens. This positions it as a more affordable alternative to OpenAI’s high-tier models. In comparison, the entry-level GPT-6 Luna costs $0.10 for inputs, $0.50 for outputs, and $0.01 for cached inputs per million tokens. The GPT-6.1 Sol model is priced at $2 for inputs, $10 for outputs, and $0.10 for cached inputs, while the flagship GPT-6 Astra commands $10 for inputs, $50 for outputs, and $1 for cached inputs per million tokens.
Sanchit Vir Gogia, an analyst at Greyhound Research, notes that while cheaper inference is a significant factor, it is only one component of agent economics. “Token processing is fuel with a higher operating cost,” Gogia said. “A CFO relieved by cheaper fuel needs to establish what happens to acceptance and review demand. In the illustrative coding workload, 5.4 additional review seconds erase the inference saving.”
OpenAI claims to have made significant investments in reducing hallucinations and factual errors produced by GPT-6.1 Terra. According to the company’s internal benchmark of error-inducing prompts, the error rate at low reasoning effort decreased from 53.6% with GPT-5.6 Terra to 16.7% with GPT-6.1 Terra. At maximum effort, the error rate fell by 18 percentage points to 4.3%, nearly matching the rates for GPT-6 Astra and GPT-6.1 Sol. Additionally, GPT-6.1 Terra meets the Critical threshold under OpenAI’s Preparedness Framework and retains the safeguards developed for GPT-6 Astra.
