GPT-6.1 Terra cuts OpenAI's mid-tier price by half

OpenAI has cut the per-token price of its mid-tier model by half and says the upgrade is faster too. GPT-6.1 Terra upgrades the model from GPT-5.6 and is available now in ChatGPT Work, Codex and the API as gpt-6.1-terra. In ChatGPT, it is rolling out to Plus, Pro, Enterprise and Edu users. Terra is the second of four tiers, from lowest to highest capability: Luna, Terra, Sol and Astra. It was left out when OpenAI launched GPT-6 Sol and Luna the previous month.

The API price is $1 per million input tokens and $5 per million output tokens. OpenAI says that is 50% lower than both GPT-5.6 Terra and GPT-6.1 Sol, as reported by BleepingComputer. 9to5Mac compares the price only with GPT-5.6 Terra.

OpenAI says Terra completes tasks more than three times as quickly as GPT-6 Astra and GPT-6.1 Sol. On Agents' Last Exam, Terra scored 15 percentage points higher than GPT-5.6 Terra at medium reasoning effort, at one-third the cost. On Terminal-Bench 4.0 at maximum reasoning effort, it scored 50.3%, against 21.5% for GPT-5.6 Terra. In its announcement, OpenAI said Terra "performs on par with GPT-6 Astra at only one-tenth its price per token".

On OpenAI's internal benchmark of error-inducing prompts, the error rate at low reasoning effort fell from 53.6% with GPT-5.6 Terra to 16.7%. At maximum effort, it fell by 18 points to 4.3%. These figures come from OpenAI's own benchmark, not an independent test. Artificial Analysis separately reported a large drop in hallucination rate on its own benchmark. For the most complex coding tasks, OpenAI still recommends GPT-6.1 Sol or GPT-6 Astra.

The price gap is the commercial point. Terra gives teams running agent work a cheaper route for tasks that do not need the top tiers. At one-tenth of Astra's price per token, OpenAI's parity claim makes Astra's premium harder to justify for any task Terra can handle. OpenAI still reserves its hardest coding work for Sol and Astra, so those tiers keep that ground.

Watch for independent testing of the error-rate gains, which currently rest on OpenAI's own benchmark. The Artificial Analysis hallucination result is the other figure to track as outside data arrives.