Theo Dufort

PurposeChoose the main goal of your AI app. This controls which cost components (model, embeddings, vector store) are included in the estimate.

Answer using your documents; needs model, embeddings, and vector store.

Conversations per monthNumber of conversations (or requests) your application handles per month. Token counts below are per conversation; total cost = tokens × conversations.

Token counts in the Model section are per conversation.

Model (completion)LLM used for generating responses. Cost is based on input (prompt) and output (completion) token volume. Prices from llm-prices.com.

Prices from llm-prices.com (updated 2025-12-17)

ModelEach provider and model has different pricing per million tokens. Output tokens are usually more expensive than input.

Tokenization (GPT-4o Mini (OpenAI))Approximate for English text. Token counts vary by language and format. Use your provider's tokenizer for exact counts.

~1.25 tokens per word · ~0.80 words per token

Input tokens (per conversation)Prompt tokens per conversation (system + user messages, context, RAG chunks). Multiplied by conversations/month for total cost.

Input words (approx.)Approximate word count for input. Editing this updates the token count using the model's conversion rate.

Cached input tokensPrompt tokens that are cached (e.g. system prompt, RAG context). Billed at a lower rate than regular input. Cannot exceed input tokens.

Output tokens (per conversation)Completion tokens per conversation. Usually billed at a higher rate than input. Multiplied by conversations/month for total cost.

Output words (approx.)Approximate word count for output. Editing this updates the token count using the model's conversion rate.

RAG retrievalConfigure how many chunks are retrieved from your vector store and added to the LLM prompt. These tokens are added to input tokens per conversation.

Top chunks retrievedNumber of most relevant chunks retrieved from the vector store per query. Common values: 3-10 chunks.

Avg. chunk size (tokens)Average number of tokens per chunk. Typical chunk sizes: 256-1024 tokens depending on your chunking strategy.

Avg. chunk size (words)Approximate words per chunk. Editing this updates the token count using the embedding model's conversion rate.

RAG context added:

2,500 tokens/conversation

Total input per conversation: 3,500 tokens (base: 1,000 + RAG: 2,500)

EmbeddingsVector representations of text for similarity search and RAG. You pay per token embedded; re-embedding is only needed when content changes.

ProviderEmbedding APIs have different dimensions and pricing. Options marked “(simulation)” use estimated costs where no public per-token price is available.

Word ↔ token conversionApproximate for English. 1 token is roughly 0.77 words. Enter volume in either field; the other updates automatically. Billing is per token.

~1.3 tokens per word · ~0.77 words per token

Tokens to embed (per month)Total tokens in the text you embed. Change this or the words field below; they stay in sync.

Words (approx.)Approximate word count. Editing this updates the token count above using the conversion rate.

Vector storeDatabase that stores embeddings and runs similarity search. Cost is typically per GB of vectors stored per month.

ProviderManaged vector DBs differ in pricing, limits, and features. Self-hosted options (e.g. pgvector) have infra cost instead of per-GB.

Storage (GB)Total size of stored vectors. Depends on dimension and number of vectors. Roughly: 1M vectors × 1536 dims (float32) ≈ 6 GB.

EstimateSum of model, embeddings, and vector store costs for the selected period. Prices are indicative; check provider docs for current rates. Base prices are in USD; other currencies use approximate exchange rates.

Model$0.83

Embeddings$0.10

Vector store$0.96

Total (month)$1.88

Model (completion)?LLM used for generating responses. Cost is based on input (prompt) and output (completion) token volume. Prices from llm-prices.com.

RAG retrieval?Configure how many chunks are retrieved from your vector store and added to the LLM prompt. These tokens are added to input tokens per conversation.

Embeddings?Vector representations of text for similarity search and RAG. You pay per token embedded; re-embedding is only needed when content changes.

Vector store?Database that stores embeddings and runs similarity search. Cost is typically per GB of vectors stored per month.

Estimate?Sum of model, embeddings, and vector store costs for the selected period. Prices are indicative; check provider docs for current rates. Base prices are in USD; other currencies use approximate exchange rates.

Model (completion)LLM used for generating responses. Cost is based on input (prompt) and output (completion) token volume. Prices from llm-prices.com.

RAG retrievalConfigure how many chunks are retrieved from your vector store and added to the LLM prompt. These tokens are added to input tokens per conversation.

EmbeddingsVector representations of text for similarity search and RAG. You pay per token embedded; re-embedding is only needed when content changes.

Vector storeDatabase that stores embeddings and runs similarity search. Cost is typically per GB of vectors stored per month.

EstimateSum of model, embeddings, and vector store costs for the selected period. Prices are indicative; check provider docs for current rates. Base prices are in USD; other currencies use approximate exchange rates.