In the AI race, the winning move isn't always a bigger model. Sometimes it's doing the same job faster and at a lower cost. OpenAI has now officially introduced GPT-6.1 Terra, a mid-tier model built on exactly that logic.

The new model slots into the company's lineup between Sol and Luna. Its specialty is routine desktop work: reading what's on screen, switching between apps, clicking around, and drafting automated responses.

Half the cost, triple the speed

For developers, the pricing is the headline. OpenAI says Terra's standard API rates are 50% lower than its predecessor, GPT-5.6 Terra. Speed gets a similar boost: according to the company's figures, the model completes computer-use tasks on average more than three times faster than GPT-6 Astra and GPT-6.1 Sol.

Hallucinations drop sharply

Confidently wrong answers remain one of users' biggest complaints, and OpenAI claims it has significantly curbed the problem in Terra. In internal tests using deliberately misleading prompts, the factual error rate at low reasoning effort fell from 53.6% in GPT-5.6 Terra to 16.7% in the new model.

With reasoning pushed to its maximum setting, the error rate drops another 18 points to 4.3%.

k

Nearly double the score in terminal benchmarks

The model's practical skills were also measured at the command line. In Terminal-Bench 4.0, which tracks how successfully coding agents execute terminal commands, GPT-6.1 Terra running at maximum reasoning effort completed 50.3% of tasks. That's more than double its predecessor's 21.5% success rate.

Worth noting: all of these figures come from OpenAI's own internal measurements and have not yet been verified by independent third-party testing.

Who can use it?

GPT-6.1 Terra is now available on ChatGPT Work and Codex for users on the Plus, Pro, Enterprise, and Edu tiers.