GPT-6.1 Terra Cuts Token Rates for Routine Agent Work

2026-10-27

Maestro’s take. Put GPT-6.1 Terra into a trial for one routine agent job. The published token rates make that worthwhile. Promote it only if the finished work costs less at the quality you need.

TL;DR

OpenAI’s GPT-6.1 Terra announcement introduces a model aimed at visual reasoning, computer use and everyday work. Standard API rates are 50% below GPT-5.6 Terra and GPT-6.1 Sol.

The model documentation lists these standard prices per million tokens:

Those are not universal rates. Above 272,000 input tokens, input and cache rates double and output rises by 50% for the full request. Cache writes and tool use can add charges.

GitHub’s October 26 changelog confirms availability in Copilot, with a gradual rollout.

Why Maestro users care

OpenAI reports better coding and computer-use benchmark results than its previous Terra. Those are vendor evaluations. They do not establish savings on your repository or workflow.

A cheaper token can still produce an expensive job if the agent needs more attempts, longer reasoning or corrective work. Maestro would start with a bounded task whose result is easy to check, such as a small code fix with a clear test.

One thing to try

Run the same representative tasks through Terra and your current model. Keep tools, permissions and acceptance checks the same. Record successful completions, total billed usage, elapsed time and rework. Choose by cost per accepted result.

Published October 27, 2026 · Brief and Maestro commentary · Tags: models, costs, agents

Original sources: OpenAI announcement, current API rates and limits, and Copilot availability.