OpenAI releases GPT-6.1 Terra and lists a lower API price

OpenAI is introducing GPT-6.1 Terra, a model it describes as fast and efficient for visual reasoning, computer use, and writing. The system-card addendum, published October 26, 2026, says the introduction is that day’s news. The product post lists standard API prices of $1 per million input tokens, $0.05 per million cached input tokens, and $5 per million output tokens, and says those standard prices are 50 percent lower than the prices of GPT-5.6 Terra and GPT-6.1 Sol. The figures and benchmark scores below are OpenAI’s measurements. They are not an independent test, and they are not a recommendation to buy access.

Where it is available

The product post says that, starting the day of the announcement, GPT-6.1 Terra is available to Plus, Pro, Enterprise, and Edu users in ChatGPT Work and Codex, and to developers in the API as gpt-6.1-terra. OpenAI’s model release notes say the same prices, and they add a limit: in ChatGPT, Terra is available in Work and Codex, not in Chat. The notes also say the model does not support UAE inference residency, and that this limit is based on where inference runs, not on the user’s location. API use is billed separately from ChatGPT subscriptions.

Amazon Web Services said on October 26 that GPT-6.1 Terra is generally available on Amazon Bedrock. AWS says customers can call it through Bedrock APIs or point ChatGPT Work and Codex at Bedrock. AWS repeats several of OpenAI’s comparison claims and says the model accepts text and image inputs. Pricing and regions, AWS says, are in the Bedrock documentation.

What OpenAI reports on its own tests

OpenAI says it measured computer use on a benchmark it calls Agents’ Last Exam. At medium reasoning effort, it says Terra scores 15 percentage points higher than GPT-5.6 Terra at about one-third the cost, performs on par with GPT-6 Astra at one-tenth Astra’s price per token, and at maximum effort is about nine times cheaper per task than Claude Opus 5.5 while finishing tasks more than three times faster than GPT-6 Astra and GPT-6.1 Sol. On MMMU-Pro, a visual-reasoning benchmark, OpenAI says Terra beats GPT-5.6 Terra at every reasoning setting it tried and nearly matches GPT-6 Astra and GPT-6.1 Sol at lower cost. On OSWorld 2.1, it says medium-effort Terra surpasses GPT-5.6 Terra’s best score at a fraction of the cost per task.

On an internal set of prompts meant to induce factual errors, OpenAI says the error rate at low reasoning effort fell from 53.6 percent with GPT-5.6 Terra to 16.7 percent with GPT-6.1 Terra, and that at maximum effort the rate fell by 18 percentage points to 4.3 percent. On Terminal-Bench 4.0, it says maximum-effort Terra solves 50.3 percent of tasks, against a published GPT-5.6 Terra score of 21.5 percent, and that GPT-6.1 Sol and GPT-6 Astra remain the better choice for the most complex coding tasks. On Terminal-Bench-Science 0.1, it says maximum-effort Terra surpasses GPT-6.1 Sol and comes within five points of Opus 5.5 at less than one-fifth the estimated cost.

The safety rating OpenAI assigns

The addendum says OpenAI is treating GPT-6.1 Terra as Critical in cybersecurity and High for biological and chemical capability under its Preparedness Framework. On that basis, OpenAI says the model uses the same safeguards stack as GPT-6.1 Sol and GPT-6 Astra, with probes updated for the new model. A footnote says the evaluations were run in a research environment or through the API, which may differ from production ChatGPT because of system prompts, tools, and effort settings. The product post adds that, on OpenAI’s evaluations, Terra is more likely than GPT-5.6 Terra to stay within scope and to disclose limitations, and that it showed strong resistance in prompt-injection tests. Those sentences are OpenAI’s assessment. This report is not safety, health, or purchasing advice.

Why it matters (analysis)

Terra is a price-and-speed model inside a family OpenAI already sells, with a narrower ChatGPT surface than the API. The same-day Bedrock listing is a distribution fact from AWS, not a second set of measurements. For the research claims OpenAI published the same day, see the three proofs it has asked outsiders to check. Both sit in the OpenAI file.

This report is not legal, financial, or health advice. See the disclaimer.