OpenAI Unveils GPT-6.1 Terra: Faster, Smarter Computer-Use AI at Half the Price

OpenAI has officially launched GPT-6.1 Terra, the latest iteration of the mid-tier model in its artificial intelligence lineup. Released on Monday, the new model arrives with a significant promise for developers and enterprise customers alike: a 50% reduction in standard API pricing compared to its predecessor, the GPT-5.6 Terra, alongside marked improvements in speed and accuracy for handling complex computer-based workflows.

Designed primarily as an everyday workhorse model, Terra is built specifically to handle routine office and digital tasks. This capability—commonly referred to in the tech industry as "computer use"—allows the AI to effectively perceive a user’s screen layout, navigate across various desktop applications, click through interface elements, and autonomously draft responses or complete administrative routines much like a human operator would.

The price cut applies directly to standard API rates, which are the fees developers pay to integrate OpenAI’s models into their own proprietary applications and enterprise software suites. In its official announcement, OpenAI highlighted the release as a major step forward. The company stated that GPT-6.1 Terra is rolling out immediately and is broadly available to Plus, Pro, Enterprise, and Edu users operating within ChatGPT Work and the Codex development environment.

ChatGPT Work serves as OpenAI’s advanced agent mode embedded inside the ChatGPT interface. It possesses the capability to operate seamlessly across multiple applications, browse the web, and execute tasks autonomously on behalf of the user. Meanwhile, Codex continues to serve as OpenAI’s specialized coding tool. For users on free-tier accounts, OpenAI maintains access through the lower-tier GPT-5.6 Luna model.

Beyond pricing and accessibility, OpenAI emphasized substantial gains in reliability and technical performance. According to internal benchmarks developed by the company using adversarial prompts—specifically engineered to trick models into making mistakes—GPT-6.1 Terra demonstrates a dramatically lower rate of factual errors. When operating at a low reasoning effort setting, which controls the duration of the model’s internal processing before responding, the error rate dropped sharply from 53.6% in the previous GPT-5.6 Terra model down to 16.7%. When pushed to its maximum reasoning effort, the error rate fell an additional 18 points, landing at just 4.3%.

Performance metrics on rigorous technical benchmarks further underscore the upgrade. On Terminal-Bench 4.0, a standardized evaluation framework that tests coding agents on tasks executed directly through a computer’s command-line terminal, Terra performed significantly better than its predecessor. Measured by the percentage of tasks completed successfully, Terra at maximum effort solved 50.3% of the assigned tasks, a steep increase compared to the 21.5% success rate recorded by GPT-5.6 Terra.

Additionally, OpenAI claims that the new model finishes designated tasks more than three times faster on average than other stable offerings in its ecosystem, such as GPT-6 Astra and GPT-6.1 Sol. Industry observers note that all of these performance figures and error tests are derived from OpenAI’s internal benchmarks rather than independent third-party audits.

The evolution of the Terra model line traces back several months through OpenAI’s fast-paced release schedule. Terra was initially introduced on June 26 as part of the broader GPT-5.6 family through a limited preview requested by the Donald Trump administration. It was subsequently opened to the general public on July 9, positioned as a model capable of competing with GPT-5.6’s predecessor generation at roughly half the cost.

As OpenAI continued to iterate on its product ecosystem, the company launched GPT-6 Sol and Luna in September, rolling out those models mere minutes after competitor Anthropic released Claude Opus 5.5. Notably, that September launch did not include an updated version of the Terra model.

While GPT-6.1 Terra is optimized for high-speed, cost-effective routine office automation and daily digital tasks, OpenAI maintains that users seeking specialized, in-depth research capabilities or complex software development assistance should continue to rely on the higher-end GPT-6.1 Sol or GPT-6 Astra models.