OpenAI releases GPT-6.1 Terra with a claimed 50% price cut over its predecessor

OpenAI released GPT-6.1 Terra on October 26, bringing its medium-sized model to ChatGPT Work, Codex and the API. The company says pricing is 50% lower than GPT-5.6 Terra’s, alongside improvements in computer use, visual reasoning and writing. The release pairs a lower price with a claim of fewer errors on OpenAI’s internal evaluation.

Terra catches up with the newer model family

Terra had been left out when OpenAI updated Sol and Luna to GPT-6 last month. This release moves Terra directly from GPT-5.6 to GPT-6.1, filling that gap rather than introducing a new model class.

OpenAI’s four main classes run from Luna at the lowest price and capability, through Terra and Sol, to Astra at the top. Terra therefore remains a middle option, below the company’s two more capable classes rather than replacing them.

The pitch spans documents, drawings and software

OpenAI describes three areas of improvement: reasoning about visual material, operating computers and producing clearer, more natural writing. Its examples stretch from drafting proposals and interpreting technical drawings to filling out forms and editing videos. The intended work covers both creating content and acting within software, rather than writing alone.

At maximum effort, a closer match to higher tiers

At maximum reasoning effort on the same benchmark, OpenAI reported an error rate of 4.3%, an improvement of 18 percentage points over GPT-5.6 Terra. The company says that nearly matches GPT-6 Astra and GPT-6.1 Sol, putting Terra closer to its higher-tier models on this particular evaluation.

These are company-run test results, not general workplace error rates. They measure responses to error-inducing prompts, rather than success across every proposal, drawing or computer task in OpenAI’s workplace pitch.