Terra's token prices put workload mix at the center of the AI bill
OpenAI's new model has different rates for input, cached input and output. A simple example shows why one headline price cannot describe a whole workload.
What does a lower AI price mean for a business that pays to process documents or automate office work? With GPT-6.1 Terra, the useful starting point is to count what the workload sends, reuses and generates.
OpenAI's launch announcement lists standard API rates of $1 per million input tokens, $0.05 per million cached input tokens and $5 per million output tokens. Its dated system-card introduction establishes the October 26 launch day. These are published rates, not a measured customer bill.
Hold the work constant
Consider a hypothetical workload with one million input tokens and 100,000 output tokens. Using uncached input, the listed token charges total $1.50: $1 for input and $0.50 for output. If all that input instead qualified for the cached-input rate, the same token counts would total $0.55.
That is an illustration of the rate structure, not a claim that every input can be cached. It also excludes any separate costs outside those token charges. A real comparison needs the quantities actually billed in each category.
Output mix matters, too. At the published rates, one million output tokens cost five times as much as one million uncached input tokens. A business generating long documents can therefore face a different cost pattern from one mostly reading them, even when both use the same model.
Then count accepted results
For investors assessing AI software economics, the next denominator is useful work completed. A task that needs another attempt, extra review or a different model can consume more resources than the first response suggests.
A practical comparison would hold the task and acceptance standard constant, record the full token mix across attempts, and include review and other delivery costs. Only then could it show whether the cost of an accepted result improved.
The launch supplies a concrete price schedule. It does not establish a customer's realized savings, adoption or profit margin. Those outcomes depend on how the model is used and how much of its output the customer can actually accept.