AI model prices keep falling, and this week OpenAI launched GPT-6.1 Terra, a mid-tier model it says is 50% cheaper than GPT-5.6 Terra and GPT-6.1 Sol. If you are planning a website chatbot, a lead-qualification agent or a tool that reads invoices, the practical question is simple: what will it cost to run, and does a cheaper model change my plan?
The prices, from OpenAI's own page
OpenAI's pricing page lists these standard rates per million tokens (a token is a small chunk of text, roughly a short word or part of one):
| Model | Input | Output | Cached input |
|---|---|---|---|
| gpt-6-astra | $10.00 | $50.00 | $1.00 |
| gpt-6.1-sol | $2.00 | $10.00 | $0.10 |
| gpt-6.1-terra | $1.00 | $5.00 | $0.05 |
| gpt-6-luna | $0.10 | $0.50 | $0.01 |
Batch processing is priced at half of the standard input and output rates for these models. Regional processing (data residency) endpoints add a 10% uplift for models released on or after 5 March 2026. Built-in tools are billed on top, for example web search at $10 per 1,000 calls plus the tokens it retrieves.
OpenAI describes Terra as suited to high-volume work such as reading incoming invoices and entering their details into accounting software. On its own internal benchmark of error-inducing prompts, OpenAI reports the error rate at low reasoning effort fell from 53.6% with GPT-5.6 Terra to 16.7% with GPT-6.1 Terra. These are the vendor's numbers on its own tests, so check them on your own tasks.
A worked example (our assumption, not a quote)
Suppose a website chatbot handles 1,000 conversations a month. Assume each one sends about 2,000 tokens in (the question, your instructions and some knowledge-base text) and gets about 500 tokens back. That is 2 million input tokens and 0.5 million output tokens a month.
| Model | Input cost | Output cost | Monthly model fee |
|---|---|---|---|
| gpt-6-luna | $0.20 | $0.25 | $0.45 |
| gpt-6.1-terra | $2.00 | $2.50 | $4.50 |
| gpt-6.1-sol | $4.00 | $5.00 | $9.00 |
| gpt-6-astra | $20.00 | $25.00 | $45.00 |
This is simple arithmetic on the published rates and ignores caching, tools and reasoning tokens, so real bills will differ. The point is the scale: at small-business volumes the model fee can be a rounding error next to development and support.
So what should a business owner do?
1. Budget for the work around the model. Build, connecting your CRM or WhatsApp, testing, monitoring and handing off to a human are usually the larger costs. Our AI chatbot development cost guide for India breaks these down.
2. Choose by accuracy first. A model that is 10 times cheaper is not a saving if it gives wrong prices or policies. Collect 30 to 50 real customer questions and score two or three models on them.
3. Match the model to the job. Simple routing and FAQ answers may suit a cheaper tier; complex multi-step agents may need a stronger one. Read AI agents for business explained before giving any agent write access to your systems.
4. Keep the model swappable. Prices and models change within months. Ask your developer to keep the model behind a settings layer so you can switch without a rebuild.
5. Set a spending cap and alerts on whatever provider account you use.
6. Protect customer data. Check where data is processed and whether regional processing (with its extra fee) matters for you.
Checklist before you commission an AI feature
- One clear task with a measurable result (for example, resolved without a human).
- Monthly volume estimate and tokens per conversation.
- Two or three models tested on your own questions.
- Estimated monthly model fee plus tool fees.
- Human approval for sensitive actions such as refunds.
- Spending cap and monitoring in place.
If you want help scoping an AI chatbot or automation with a clear monthly cost, see our AI and automation service or contact us.
