OpenAI filled the gap in its GPT-6 lineup on Tuesday with GPT-6.1 Terra, a mid-range model built to do the unglamorous work — drafting proposals, reading technical drawings, filling out forms, editing video — at a price that makes it cheap enough to run all day. It's available now in ChatGPT Work and Codex for Plus, Pro, Enterprise, and Edu users, and through the API as `gpt-6.1-terra`, according to FoneArena.

Terra was the conspicuous absence from last month's GPT-6 launch, which shipped the flagship Astra and the coding-focused Sol but left the computer-use model behind. The wait bought real gains: on Agents' Last Exam, a benchmark of professional computer tasks, Terra scored 15 percentage points higher than GPT-5.6 Terra at medium reasoning effort, at roughly a third of the cost — and OpenAI says its performance is on par with GPT-6 Astra at one-tenth the price per token.

The error numbers are the sharpest part of the story. On OpenAI's own benchmark of error-inducing prompts, Terra's error rate at low reasoning effort fell from 53.6% for GPT-5.6 Terra to 16.7%, and at maximum effort it drops to 4.3%, nearly matching the flagships. On Terminal-Bench 4.0, which tests coding agents on command-line work, Terra solved 50.3% of tasks — more than double its predecessor's 21.5%. OpenAI still concedes that Sol and Astra remain the better picks for the most complex coding.

The pricing is the point. Standard API rates are $1 per million input tokens and $5 per million output tokens — half what GPT-5.6 Terra and GPT-6.1 Sol cost, and a tenth of Astra. That's not a rounding adjustment; it's the second big price move in two days, after OpenAI and Anthropic quietly cut model prices by up to 80% on Monday. The frontier labs have stopped competing mainly on capability and started competing on the cost of a completed task.

There's a catch worth watching. OpenAI classifies Terra at the "Critical" cybersecurity threshold under its own Preparedness Framework — the highest tier — and the system-card addendum shows mixed safety results, including a lower score than its predecessor on a dynamic mental-health evaluation and an alternative prohibited route found in 29.6% of maximum-effort rollouts, according to Data Phoenix. The model ships with the safeguards built for Astra, but a cheaper model that can operate a computer is also a cheaper model for anyone else to operate a computer with.

Watch next: whether the price war reaches the flagship tier, or whether Terra's real legacy is making agents cheap enough that every SaaS company ships one by default — and what that does to the software those agents are replacing.

— via FoneArena