GPT-6.1 Terra halves OpenAI's mid-tier API price

OpenAI released GPT-6.1 Terra on October 26, 2026, a mid-priced model that costs $1 per million input tokens and $5 per million output tokens. That is half the price of GPT-5.6 Terra, the model it replaces. The new model also adds computer use, so agents built on OpenAI's cheaper tier can now operate a screen directly.

A token is a small chunk of text, roughly part of a word. API prices are quoted per million of them.

Where Terra sits in OpenAI's lineup

OpenAI sells four tiers of model. 9to5Mac lists them as Luna, Terra, Sol and Astra, with Terra in the middle. Terra was left out when OpenAI shipped GPT-6 Sol and Luna in September 2026, so this release brings the mid tier up to date.

Key facts from the three reports:

Detail	GPT-6.1 Terra
API model name	gpt-6.1-terra
Input price	$1 per million tokens
Output price	$5 per million tokens
Replaces	GPT-5.6 Terra
ChatGPT access	Plus, Pro, Enterprise and Edu, in ChatGPT Work
Other access	Codex and the API

All three outlets put the cut at 50 percent against GPT-5.6 Terra. BleepingComputer also says it is 50 percent cheaper than GPT-6.1 Sol. The other two do not make that comparison.

What computer use adds

Computer use means the model can read what is on a screen, click, type and switch between apps. The Next Web reports that Terra can now do this. It also reads images better, for example turning architectural drawings into 3D models.

OpenAI's own description, as quoted by BleepingComputer, calls it "a fast, efficient model that combines detailed visual reasoning and precise computer use." The Next Web adds that Terra is more likely to stay within its assigned task and to say what it cannot do.

OpenAI's benchmark claims

Every figure below comes from OpenAI. BleepingComputer says it did not verify them.

Test	GPT-5.6 Terra	GPT-6.1 Terra
Terminal-Bench 4.0, maximum effort	21.5%	50.3%
Internal error rate, low effort	53.6%	16.7%
Internal error rate, maximum effort	not reported	4.3%

Terminal-Bench measures how many command-line coding tasks a model completes. The error-rate test is an internal OpenAI set of prompts designed to make models state wrong facts. At maximum effort, the error rate fell by 18 percentage points.

On Agents' Last Exam at medium effort, BleepingComputer reports a 15-point gain over GPT-5.6 Terra at about one-third the cost. OpenAI also says Terra finishes tasks more than three times faster than GPT-6 Astra and GPT-6.1 Sol. "It even performs on par with GPT-6 Astra at only one-tenth its price per token," OpenAI says.

OpenAI still recommends Sol or Astra for the hardest coding work, according to both BleepingComputer and The Next Web.

The safety rating

The Next Web reports that OpenAI rates Terra "Critical" for cybersecurity skills under its Preparedness Framework. That is the same level GPT-6 Astra reached first, and Terra ships with the same protections. The release follows OpenAI's decision last month to shelve GPT-6.1 Astra after failed internal safety tests.

What this means for developers

Terra is now the obvious default for high-volume work that does not need OpenAI's top models.

• Switch the model name and re-run your evals. Replace GPT-5.6 Terra calls with gpt-6.1-terra on a sample of real traffic. Compare quality before moving everything.
• Recheck your budget. At $1 and $5 per million tokens, the same workload costs half what it did on GPT-5.6 Terra.
• Try computer use on cheap tasks first. Form filling, data entry and app-to-app copying are good tests. Keep a human check on anything that sends money or messages.
• Keep Sol or Astra for hard code. OpenAI itself says the bigger models still win on the most complex coding tasks.
• Expect cyber safeguards. A "Critical" rating means security-related requests may meet the same limits as Astra. Test any security tooling before you depend on it.

The benchmark gains are large but self-reported. Watch for independent results on Terminal-Bench and on agent tests in the coming weeks.