Embedding Cost Calculator

An embedding cost calculator estimates the API expense of converting text documents into vector embeddings — the numerical representations that power semantic search, RAG systems, and AI-powered retrieval. Unlike LLM inference costs (which are charged per input + output token), embedding APIs charge only for input tokens — the text you send to be embedded. Our embedding cost calculator online models all three cost phases of a real production pipeline: the one-time initial corpus ingest, ongoing monthly re-embedding as your knowledge base grows, and per-query embedding costs at search runtime. Everything runs in your browser — no data leaves your device.

1. Choose a Dataset Size Preset or go Custom: Select from preset sizes (Small KB ~500 docs, Medium KB ~10K, Large KB ~100K, Enterprise ~1M) or enter your exact document count and average token count per chunk. The corpus summary updates instantly.
2. Select an Embedding Model: Pick from 11 models across OpenAI, Google, Cohere, Voyage AI, and Mistral. The model selector shows cost per million tokens, output dimensions, and maximum context window. Models with higher dimensions produce more accurate embeddings but use more storage.
3. Set Monthly Refresh and Query Volume: Enter what percentage of your corpus is re-embedded each month (e.g. 5% for a slowly growing knowledge base) and how many search queries trigger a query embedding per day. The calculator separates these into distinct cost components.
4. Read the Results: The calculator shows one-time ingest cost, monthly ongoing cost, first-year total, and per-component breakdown. The cross-model table sorts all 11 models by first-year total for your exact configuration. Click any row to switch models.

• Corpus Size & Chunk Tokens: The dominant cost driver for most projects is the initial ingest — embedding every chunk once. A 100K-chunk corpus at 600 tokens each is 60M tokens. At $0.020/1M (text-embedding-3-small), that is $1.20. At $0.130/1M (text-embedding-3-large), that is $7.80. Choose your model based on quality needs, not just unit price.
• Corpus Refresh Rate: A living knowledge base requires re-embedding updated documents. A 5% monthly refresh on 1M chunks adds 50K re-embedded chunks per month — an ongoing cost that compounds if you have a high churn corpus like a product catalog or news archive.
• Query-Time Embedding: Every semantic search query requires embedding the query text. At high QPS (queries per second), query embedding costs can rival or exceed corpus refresh costs. A 50-token query at 500 queries/day is 25K tokens/day — negligible alone, but significant at scale.
• Vector Storage Dimensions: Higher-dimension models (3072d) require 4× the storage of 768d models. For 1M vectors: text-embedding-3-large needs ~12 GB of raw float32 storage vs. ~3 GB for text-embedding-004. Storage costs in hosted vector databases can exceed embedding API costs for large deployments.

The highest-ROI optimization for most embedding pipelines is model selection: OpenAI's text-embedding-3-small at $0.020/1M and Voyage AI's voyage-3-lite at $0.020/1M deliver excellent retrieval quality at a fraction of the cost of premium models. For English-language RAG, these are often indistinguishable from more expensive options in production retrieval benchmarks. Additionally, chunk size matters — embedding 512-token chunks instead of 1024-token chunks halves your token count and storage footprint with minimal retrieval quality impact. Finally, implement an embedding cache to avoid re-embedding identical query strings — for chatbots and support portals with repetitive queries, this alone can cut query embedding costs by 60–80%.