What will an AI feature cost you every month?
Not to build: to run. Tell us what it does and how often, and see the same feature priced across eight models in naira, with the cheap and the expensive way to do it side by side.
Every month, at that volume
₦3,139 to ₦78,469
Depending entirely on which model answers it. Prices reviewed September 2026; converted at ₦1,550 to the dollar.
That is 25x between the cheapest model and the dearest, for identical work. Which one answers is worth more than any amount of prompt tuning, and most features do not need the dearest.
Model by model
- GPT-5.6 LunaOpenAI₦3,13963 kobo eachThe cheap one. Classifying, tagging, routing, short replies.
- Gemini 3 FlashGoogle₦7,847₦1.6 eachFast and cheap, good at reading long documents.
- Claude Haiku 4.5Anthropic₦14,124₦2.8 eachSmall and quick, and still careful with instructions.
- Claude Sonnet 5Anthropic₦28,249₦5.6 eachThe workhorse. Most features we build sit here.
- Gemini 3.1 ProGoogle₦31,388₦6.3 eachThe middle of Google's range.
- GPT-5.6 TerraOpenAI₦31,388₦6.3 eachThe middle of OpenAI's range.
- Claude Opus 5Anthropic₦70,622₦14 eachThe heavy one. Worth it where the answer has to be right.
- GPT-5.6 SolOpenAI₦78,469₦16 eachOpenAI's flagship.
What this does not include
- Caching, which most providers bill at a tenth of the input rate and which a well-built feature leans on heavily. Assume the real figure is lower than the one above, not higher.
- Batching, worth about half price where the answer can wait a few hours.
- Our fee for building it. That is the other tool: what it costs to build.
- Anything a provider charges for images, audio or long-context tiers. The table above is text in, text out; check the provider’s own page before you sign anything.
Worth saying plainly
Sometimes the answer is not a model
If the job is sorting things into a fixed set of buckets, matching a record or answering from a table you already have, ordinary code does it faster, cheaper and the same way every time. We will tell you that before you pay us to build the other thing.
The bill is mostly about shape
- Writing costs several times more than reading. Output tokens run four to six times the price of input on every model. A summary is cheap; a draft from scratch is not.
- The model you pick matters more than the prompt. The spread between the cheapest and dearest model above is over twentyfold for identical work, and most features do not need the dearest.
- Sometimes the answer is ordinary code. Sorting, matching, filling in a table you already have: code does that faster and cheaper, and we will say so before quoting.
We build the ones that earn their place.
Bring the figure above and the feature you have in mind. We will tell you which model it actually needs, what it would cost to build, and whether it should be a model at all.