AI Embedding Cost Calculator

The calculator treats every embedding model the same way: take the input text in tokens, divide by 1,000,000, and multiply by the vendor's published $/M-token price. That price comes from the vendor pricing pages cited at the bottom of this page, hand-verified on 2026-05-12and stored alongside its source URL in the calculator's code.

Three line items add up to a first-year cost. The math is intentionally boring:

index_tokens         = documents × avg_tokens_per_doc
index_cost_usd       = index_tokens / 1,000,000 × price_per_million
monthly_query_tokens = queries_per_day × avg_tokens_per_query × 30
monthly_query_usd    = monthly_query_tokens / 1,000,000 × price_per_million
reindexes_year_1     = reindex_months > 0 ? floor(12 / reindex_months) - 1 : 0
first_year_usd       = index_cost_usd
                     + reindexes_year_1 × index_cost_usd
                     + 12 × monthly_query_usd
first_year_lkr       = first_year_usd × usd_to_lkr_rate

The reindex term models the realistic case where a knowledge base gets re-embedded on a schedule — content drifts, chunking strategies change, vendors ship a better model — and the initial corpus pass has to be paid again. floor(12 / reindex_months) - 1 counts the additionalreindexes in the first 12 months; quarterly is 3 extra passes, monthly is 11, yearly is 0, and 0 months means "never".

Google's Vertex AI embedding models text-embedding-005 and text-multilingual-embedding-002 publish prices per character rather than per token. The calculator normalises using Google's documented 4-chars-per-token approximation so every row compares like-for-like. The page exposes a cross-check that re-derives the per-million-token price from the published per-character figure and confirms the stored value: char→token cross-check passes ($0.10/M tokens). Where a tier doesn't publish a public price (enterprise-only contracts, fine-tuned variants) the row is omitted, not estimated.

Worked-example self-test (computed live on this page) — each line reconciles the formula above with the hand-derived numbers in the code header:

• ✓5k×4k corpus, 500q/day×50t, yearly reindex — OpenAI 3-small → expected $0.58, got $0.58
• ✓Same workload — Cohere English v3 → expected $2.90, got $2.90
• ✓50k×2k corpus, 5kq/day×80t, quarterly reindex — OpenAI 3-large → expected $70.72, got $70.72
• ✓Zero documents, zero queries — Voyage 3-lite → expected $0.00, got $0.00
• ✓Monthly reindex of 5k×4k corpus — OpenAI 3-small → expected $4.98, got $4.98

Out of scope for v1: vector-database storage (Pinecone, Weaviate, Qdrant, pgvector are a separate cost line); chat-completion / RAG generation cost (use the upcoming AI API cost calculator); token counting (use the AI Token Counter); reranker pricing; fine-tuned embedding tiers; and self-hosted GPU embeddings, which are billed by hardware-hour, not per token.

• OpenAI — API Pricing (text-embedding-3-small / 3-large / ada-002)
• Cohere — Pricing (embed-english-v3.0, multilingual-v3.0, light-v3.0)
• Voyage AI — Pricing (voyage-3, voyage-3-lite, voyage-3-large, code-3, multilingual-2)
• Google Cloud — Vertex AI Generative AI Pricing (gemini-embedding-001, text-embedding-005)
• Mistral AI — Technology pricing (mistral-embed)
• Google — Tokens documentation (4-chars-per-token approximation)
• Central Bank of Sri Lanka — Daily indicative exchange rates (USD→LKR)

Pricing was last cross-checked against each vendor's pricing page on 2026-05-12. The table is reviewed quarterly and whenever a vendor announces a tier change. Spotted a stale row? Email the address below — most fixes ship within 24 hours.