GPT-6.1 Terra is the latest in a line of upgrades to the middle-tier AI model from OpenAI. The new tool is mostly intended to speed up the tedious desktop work of clicking through web apps, filling out complex forms and editing code. Instead of forcing developers and teams to pay steep subscription fees for automated workflows, this new version makes screen-based tasks significantly cheaper to run.
Lower costs and faster screen navigation
OpenAI confirmed that standard API pricing for the new model is half of what its predecessor cost. Developers and subscribers using ChatGPT Work or Codex can jump right in today with much lower rates for processing everyday text and code input. Beyond the huge price drop, the software moves through desktop actions three times faster than heavier models in the same lineup.
On technical command-line evaluations like Terminal-Bench 4.0, the model completed over 50% of coding challenges when set to maximum effort. It also manages to match top-tier performance on multi-step software benchmarks while consuming far fewer computing tokens along the way. That extra speed boost makes routine chores like browsing websites, parsing files, and editing documents feel much smoother during busy workdays.
Fewer mistakes and smarter guessing
Internal testing from OpenAI highlights a massive drop in factual mistakes. More specifically, it cut error rates at lower effort settings from over 50% down to 16%. Independent benchmark evaluations from Artificial Analysis back up those claims, placing the model at an impressive spot for total intelligence per dollar spent. The system also resorts to wild guessing far less often when it lacks a clear answer, making automated data entry much more reliable.
When stacked against rival tools like Anthropic’s Claude Opus 5.5, this new release holds its ground on complex software assignments while costing roughly nine times less per task. That massive price gap gives developers a realistic way to run automated web agents continuously without racking up giant server bills. While heavier models still exist for deep research, this middle option easily handles the vast majority of routine office work.
Better security for automated agents
Handing over control of your mouse and keyboard always requires serious safety guardrails. So, the model inherits heavy cybersecurity protections from OpenAI’s flagship architecture. Improved defense against hidden prompt injection attacks keeps autonomous web agents operating safely within set boundaries when interacting with third-party sites. Bringing fast, reliable screen control down to budget pricing is a welcome change for anyone trying to build useful AI tools.
The post OpenAI’s New GPT-6.1 Terra Delivers Flagship Computer Control for Cheap appeared first on Android Headlines.
​Â
