Agent Upstart

Cool things people are building with agents — and practical help getting the most out of them.

Playbooks · Coding tasks

Your Work default changed to Terra: check your preset and pick Sol back for the hard jobs

Written November 1, 2026

For a paying ChatGPT user whose Work answers suddenly come from a different model: find out what changed in your default preset, confirm your limits did not change, and get Sol back for the jobs Terra is not the pick for.

What changed

OpenAI released GPT-6.1 Terra on 26 October 2026. It is rolling out in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise, and Edu plans, and in the API as gpt-6.1-terra. The rollout is gradual: it reaches accounts at different times, so you may not see the change yet. The release notes say your plan's overall usage limits are unchanged. They also say some Plus and Pro users may see GPT-6.1 Terra at Medium instead of GPT-6.1 Sol at Light in the default Power presets in Work. Standard and Fast modes are supported where available. Terra is available in Work and Codex, not in Chat, and it does not support UAE inference residency. OpenAI describes Terra as a fast, efficient model for visual reasoning, precise computer use, and clear, natural writing, and its standard API prices are $1 per million input tokens, $0.05 per million cached input tokens, and $5 per million output tokens, 50% lower than GPT-6.1 Sol. For the most complex coding tasks, OpenAI's announcement says GPT-6.1 Sol and GPT-6 Astra remain the better choice. That line from OpenAI is the only basis in this page for picking Sol back: this page does not sort tasks between the models beyond it.

Before you start

Open ChatGPT Work and look at the preset and model shown for a new task before you type anything. Write down the model name and the effort level exactly as shown, for example Terra at Medium or Sol at Light. If the preset still shows Sol at Light, the change has not reached your account yet, and the rest of this page is a check to repeat later, not a fix to apply now.

Steps

  1. Confirm which preset you are on. In Work, open the model list for a new task and read the default Power preset: the model name and the effort level. Compare it with what you wrote down before. If it shows Terra at Medium where it used to show Sol at Light, the preset change described in the release notes has reached your account. If it shows something else, go by what is shown, not by this page.
  2. Confirm your limits did not change. The release notes say overall usage limits are unchanged by the rollout. Do not treat slower or shorter answers as a limit change: a different model at a different effort level can answer differently within the same limits. If you hit a limit message, read what it says before assuming the preset caused it.
  3. Try one routine job with the new preset before judging it. Give Terra a routine job of the kind OpenAI describes it for, such as drafting, reading a screenshot or a form, or a small, easily checked edit, and read the result against what you asked for. One routine job tells you how the new preset handles routine work. It tells you nothing about the hardest coding jobs.
  4. Pick Sol back for the hardest coding jobs. For the most complex coding tasks, OpenAI says Sol and Astra remain the better choice. In Work or Codex, use the model list to select GPT-6.1 Sol for that job instead of accepting the preset. If Sol is not listed for your account, do not assume it was removed: check your plan, region, and workspace settings, which the release notes name as the conditions that decide what you can use.
  5. Check again after the rollout reaches you, and after any preset change. Because the rollout is gradual, a preset that shows Sol today can show Terra later. Repeat step 1 when answers start sounding different, or after an update to Work or Codex. The preset you are on is a fact to read off the screen each time, not a setting to assume.

What can go wrong

Limits

Related playbooks

Sources

Back to: Coding playbooks hub · Also: Try OpenAI's GPT-6.1 Terra safely: run it supervised on a throwaway task first · Try Terra in Copilot supervised first, the same way as in Codex · Pick the cheapest model for the job