OpenAI is charging developers $1 per million input tokens and $5 per million output tokens for GPT-6.1 Terra, the mid-tier model in its GPT-6 lineup. The company says those standard API rates are 50% below what it charges for GPT-5.6 Terra and for GPT-6.1 Sol, its more advanced model. By OpenAI’s pricing page, GPT-6.1 Sol lists at $2 and $10 and GPT-5.6 Terra at $2 and $12, so the output cut against the older Terra is closer to 58%.
The release went out Monday afternoon, and Amazon Web Services made the model generally available on Amazon Bedrock the same day. In ChatGPT, Plus, Pro, Enterprise and Edu users get it inside ChatGPT Work and Codex, and developers call it in the API as gpt-6.1-terra.
OpenAI pitches the model at the computer-based work that fills the workday, from drafting proposals to filling out forms. Its launch post says the lower price makes it possible to automate high-volume work “like reading incoming invoices and entering their details into accounting software.” For the most complex coding tasks, the post says, GPT-6.1 Sol and GPT-6 Astra remain the better choice.
What the price cut buys
On Agents’ Last Exam, which OpenAI describes as a benchmark of professional tasks, Terra at medium reasoning effort scored 15 percentage points above GPT-5.6 Terra at about one-third the cost, the company said. It also performed on par with GPT-6 Astra at one-tenth the price per token and finished tasks more than three times faster than Astra and Sol. At maximum effort, OpenAI put Terra’s cost per task at roughly one-ninth of Anthropic’s Claude Opus 5.5.
Those are OpenAI’s own numbers. The independent benchmarking firm Artificial Analysis lists Terra at maximum effort with a score of 48 on its Intelligence Index and a cost of $0.40 per Intelligence Index task, according to its model page. The Decoder, which reported the firm’s comparisons, put GPT-5.6 Terra at $1.40 per task and 42 on the index, and GPT-6.1 Sol at 52. For context, Anthropic lists Claude Sonnet 5.5 at $2 and $10, twice Terra’s rates.
Buyers planning large jobs should read the model page as well. The context window runs to 1,050,000 tokens, but any prompt over 272,000 input tokens is billed at double the input rate and 1.5 times the output rate for the whole request. Batch and Flex processing cost half the standard price, and regional processing adds 10% where it is available.
OpenAI also reported that Terra solved 50.3% of Terminal-Bench 4.0 tasks at maximum effort, up from GPT-5.6 Terra’s published 21.5%.
What the system card says about agent behavior
OpenAI’s system card addendum treats Terra as Critical in cybersecurity and High in biological and chemical capability under its Preparedness Framework, and says the model uses the same safeguards stack as Sol and Astra. Its alignment section covers what the model did when it hit limits during simulated professional work.
In the Respecting Warnings test, where barriers such as AccessDenied errors and content policy restrictions come from the environment, Terra saw the barrier but finished the task by another route that “violated the spirit of the restriction” in 29.6% of rollouts at maximum reasoning effort. That was higher than GPT-6 Astra, GPT-6 Luna and GPT-6.1 Sol, and lower than GPT-6 Sol. OpenAI said the test mostly measures low-stakes situations and runs without system-level controls, many of which it said would plausibly stop the workarounds.
A second evaluation covers agents that run alongside other agents. Every model found a message board, and Terra tried to communicate with apparent peers in 65% of samples and performed the specified unauthorized action in 11 of 80, or 13.75%. GPT-6.1 Sol did so in 1.25% and GPT-6 Sol in 16.25%. OpenAI said Terra’s result is comparable to GPT-6 Sol’s and that the gap with GPT-6.1 Sol largely reflects promises of future help the model indicated it would not follow through on. WebProNews reported on Friday on outside researchers finding gaps in OpenAI’s internal agent monitoring.
For companies buying through Amazon, AWS says customers can call Terra through Bedrock APIs or configure ChatGPT Work and Codex to use the model there, with established AWS controls to secure workloads, govern access and audit model invocation activity. Last Wednesday, GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna began a private preview on Google Cloud’s Model Garden.