Build My AI Stack
Pick a use case and your constraints. Each stage gets a model from the live catalog with the reason it was chosen, and you get a monthly cost estimate for the whole stack.
Estimated model cost for support assistant
$4 / month
50,000 requests a month, 2,000 input and 300 output tokens each, using the suggested generation and routing models. Excludes embeddings, reranking, hosting and engineering.
1. Intent & routing
Classify each request, apply access and policy checks, and route it.
Cheapest listing that reports structured output, so it can return a category reliably. Prompts here are short, so price matters more than size.
2. Retrieval
Embed documents and queries so only relevant, permitted content reaches the model.
Lowest-priced embedding model. Pick once and keep it: changing embedding models means re-indexing everything.
3. Reranking
Reorder retrieved passages so the best evidence comes first.
No priced reranking model in the catalog yet; many teams start without one.
4. Generation
Produce the answer, draft or action.
- gpt-oss-20b · OpenAISuggested
Lowest blended price among listings with tool calling, at least 32K context. $0.02 in / $0.09 out per 1M.
Option 2 among listings with tool calling, at least 32K context. $0.04 in / $0.04 out per 1M.
Option 3 among listings with tool calling, at least 32K context. $0.03 in / $0.13 out per 1M.
Cheapest is not best: evaluate the top options on your own test set before choosing.
5. Validation & response
Check format, citations and safety before anything reaches the user; log for observability.
The same small structured-output model can act as a checker: validate the format and flag unsupported claims for review.
Suggestions are the lowest-priced model makers' API listings that report the needed capabilities in the catalog; capabilities are as reported by each source. Prices are list prices with sources on each model page. Use these as a shortlist, then compare them in Compare and test quality yourself.