OpenAI Releases GPT-6.1 Terra, a Faster AI Model With 50% Lower API Pricing

OpenAI has introduced GPT-6.1 Terra, a faster, lower-cost AI model designed for computer-based tasks, visual reasoning and software development. The model is now available through ChatGPT Work, Codex and the OpenAI API, with standard API pricing 50% below GPT-5.6 Terra and GPT-6.1 Sol. OpenAI has also added Terra to Arena.ai's Agent Arena and Code Arena platforms, allowing users to evaluate its performance on practical assignments.

Terra is designed to handle everyday professional work that involves interacting with software and interpreting visual information. OpenAI highlighted applications ranging from processing invoices and completing forms to preparing business documents, analyzing technical drawings and editing video. The model is intended to provide much of the functionality of more expensive AI systems while completing tasks faster and using fewer resources.

According to OpenAI's evaluations, GPT-6.1 Terra represents a substantial improvement over its predecessor, particularly in computer use. On Agents' Last Exam, which measures performance across complex professional assignments, Terra scored 15 percentage points higher than GPT-5.6 Terra at medium reasoning effort while costing approximately one-third as much per task.

The benchmark includes assignments such as converting architectural plans into three-dimensional models and analyzing film footage to produce detailed restoration records. OpenAI says Terra achieved performance comparable to its higher-end GPT-6 Astra model at one-tenth the price per token.

Speed is another major focus. OpenAI reports that Terra completes tasks more than three times faster than both GPT-6 Astra and GPT-6.1 Sol. At maximum reasoning effort, the company says Terra is also more than three times as token-efficient as Anthropic's Claude Opus 5.5, resulting in approximately nine times lower costs per task.

Terra also showed improvements on benchmarks measuring visual interpretation and desktop application use. On MMMU-Pro, the model surpassed GPT-5.6 Terra across all reasoning settings and approached the performance of GPT-6 Astra and GPT-6.1 Sol at a lower cost. On OSWorld 2.1, which evaluates an AI agent's ability to complete multistep assignments in desktop software, Terra exceeded its predecessor's highest reported score while requiring substantially less spending per task.

OpenAI reported gains in factual accuracy as well. On an internal evaluation designed to expose factual mistakes, GPT-6.1 Terra's error rate fell to 16.7% at low reasoning effort, compared with 53.6% for GPT-5.6 Terra. At maximum effort, the newer model recorded a 4.3% error rate, an improvement of 18 percentage points that brought it close to the results of GPT-6 Astra and GPT-6.1 Sol.

The improvements extend to programming tasks. On Terminal-Bench 4.0, which tests an AI coding agent's ability to resolve software problems through command-line tools, Terra completed 50.3% of assignments at maximum reasoning effort. Its predecessor achieved 21.5% on the same benchmark.

OpenAI nevertheless identifies GPT-6.1 Sol and GPT-6 Astra as stronger options for the most demanding software engineering assignments. Terra's advantage is its combination of speed, cost and capability rather than outright leadership across every category.

On Terminal-Bench-Science 0.1, a benchmark involving scientific research tasks performed with code, Terra exceeded GPT-6.1 Sol's performance while operating at a lower cost. It also finished within five percentage points of Claude Opus 5.5 while costing less than one-fifth as much, according to OpenAI's estimates.

Alongside those company-reported results, Terra is entering independent community evaluations through Agent Arena and Code Arena. Both platforms are operated by Arena.ai and allow users to compare models using practical tasks rather than relying exclusively on benchmarks published by their developers.

Agent Arena measures how well AI systems plan and execute assignments involving multiple steps, while Code Arena concentrates on programming performance. Agent Arena has accumulated more than 2.5 million sessions and uses causal tracing techniques in its evaluations.

Terra's participation is notable because the model had previously appeared in Microsoft's internal coding evaluations as a judge responsible for assessing results across 18 configurations. Its addition to the public arenas allows its own capabilities to be tested rather than using it solely to evaluate other systems.

GPT-6.1 Sol, another model in OpenAI's lineup, became available for community testing on the same platforms on September 29, 2026. Sol subsequently reached the top five in efficiency rankings, with reported median task costs between $0.56 and $0.82 and an 11.23% improvement in cost-performance measurements.

Terra's eventual rankings and task costs will provide additional points of comparison with Sol and other models already participating in the evaluations. Those results will also help distinguish its performance on user-submitted assignments from OpenAI's internal testing.

OpenAI has separately outlined safety measures for the new model. GPT-6.1 Terra meets the Critical cybersecurity capability threshold under the company's Preparedness Framework and retains protections developed for GPT-6 Astra. OpenAI says its evaluations also found improvements in the model's ability to follow task boundaries, communicate limitations and resist malicious instructions embedded in external content.

GPT-6.1 Terra is available to ChatGPT Plus, Pro, Enterprise and Edu subscribers through ChatGPT Work and Codex. Developers can access the model through the API using the identifier gpt-6.1-terra.

API pricing is set at $1 per million input tokens, $0.05 per million cached input tokens and $5 per million output tokens. OpenAI says existing usage allowances in ChatGPT Work and Codex will support more tasks with Terra than with GPT-5.6 Terra, GPT-6.1 Sol or GPT-6 Astra.

The release gives developers and professional users another option for automating computer-intensive work without relying on OpenAI's most expensive models. While the company's benchmarks indicate substantial improvements in speed, accuracy and efficiency, Terra's performance in Agent Arena and Code Arena will offer an additional measure of how those capabilities translate to practical tasks.

See more here:

This analysis is based on reporting from OpenAI & Crypto Briefing.

Image courtesy of OpenAI.

This article was generated with AI assistance and reviewed for accuracy and quality.