// approximate tokens, chars, and batch costs for embedding workloads
Estimate token counts, character totals, batch sizes, and API costs for AI embedding workloads across OpenAI, Cohere, and Voyage models.
Ready to estimate
Configure your workload and click Estimate CostPaste a sample document or enter character/token counts manually using the toggle.
Pick your embedding model and set batch size per API request.
See total tokens, characters, request count, and estimated USD cost instantly.
This tool calculates approximate token counts, character totals, and API costs for AI text embedding workloads. It supports the most popular providers and lets you define custom per-token pricing. Useful for RAG systems, vector databases, and semantic search pipelines.
Token counts are approximated using a standard ratio of ~4 characters per token, which is a good heuristic for English text. For precise counts use a tokenizer library like tiktoken. Our estimate is suitable for budgeting and planning.
Batch size is the number of text inputs sent in a single API call. Most providers accept up to 100–2,048 inputs per request. Larger batches reduce round-trip overhead but don't change the token-based pricing.
All major providers (OpenAI, Cohere, Voyage AI) price embeddings per token consumed. Batch calls may have minimum token billing. This estimator uses per-token pricing throughout.
Each embedding model has a maximum token limit per input document. Text exceeding this limit must be chunked. The tool warns you if your average document length approaches the model's context window.
Paste a representative sample document in text mode, then set the document count to your total. The estimator multiplies tokens-per-document by your document count to produce total tokens and cost.
For general-purpose RAG, text-embedding-3-small offers excellent quality at low cost. For higher accuracy on complex queries use text-embedding-3-large. Voyage AI models are strong for domain-specific use cases like finance or code.
Yes — completely free, browser-based, and no sign-up required. All calculations happen client-side. No text you enter is ever sent to a server.
Yes. Select "Custom model..." from the model dropdown and enter your own price per 1M tokens. This is useful for private model deployments or negotiated API contracts.
An embedding cost estimator helps developers and ML engineers predict how much it will cost to embed a text corpus using a vector embedding API. Whether you're building a retrieval-augmented generation (RAG) pipeline, a semantic search engine, or a recommendation system, understanding embedding costs before you run your workload can save you from surprise bills.
This free browser-based tool calculates approximate token counts, total character volumes, API request counts, and total USD cost for the most popular embedding providers — OpenAI, Cohere, and Voyage AI — without sending a single character of your data to any server.
💡 Looking for premium web development assets? MonsterONE offers unlimited downloads of templates, UI kits, and developer tools — worth checking out.
All major embedding APIs charge per token rather than per character or per request. Tokens are sub-word units produced by a tokenizer — BPE (byte-pair encoding) for OpenAI models, WordPiece for some others. As a rule of thumb, 1 token ≈ 4 characters for English text, or roughly 0.75 words. This tool uses that 4:1 character-to-token ratio to produce a fast, reliable estimate without requiring you to run a full tokenizer.
For precise counts on production workloads, use a library like tiktoken (OpenAI) or the provider's tokenization endpoint. But for budgeting and capacity planning, the approximation here is accurate within 5–10% for standard English prose.
Embedding prices are typically quoted per million tokens (1M tokens). As of early 2026, costs range from $0.020/1M tokens for efficient small models up to $0.130/1M tokens for high-dimensional large models. Here's a quick overview of the models included in this estimator:
When calling an embedding API, you can send multiple input texts in a single HTTP request. This is called batching. The batch size determines how many documents are grouped into one call. While batching doesn't change your token cost, it dramatically affects:
This estimator divides your total document count by batch size to show you the number of API requests required. For most production pipelines, a batch size of 100–500 documents (depending on average document length) is a practical default.
Every embedding model has a maximum input size measured in tokens. OpenAI models support up to 8,191 tokens per document. Cohere's v3 models support 512 tokens per document. Voyage AI models support a generous 32,000 tokens per input. If a document exceeds the model's context window, the API will either truncate it silently or return an error, depending on the provider settings.
This is why document chunking is a critical step in any RAG pipeline. You split long documents into smaller, overlapping chunks before embedding, ensuring every chunk fits within the model's context window. Common chunking strategies include fixed-size character splitting, sentence-aware splitting, and recursive character splitting with overlap.
A typical RAG pipeline involves three phases that incur embedding costs:
For most applications, indexing cost dominates. Use this tool to estimate your one-time indexing cost by entering your total corpus size. Then separately estimate query cost using your expected query volume.
There are several practical strategies to reduce the total cost of your embedding workload:
text-embedding-3-small or voyage-3-lite deliver excellent retrieval quality at a fraction of the cost of large models. Benchmark before committing to the most expensive option.Switch between Text Input mode (paste a sample document and set a document count) and Manual Input mode (enter total character or token counts directly). Select your embedding model from the dropdown — or choose "Custom model" to enter your own per-token price. Set your batch size, then click Estimate Cost to see the full breakdown. The output includes total tokens, total characters, number of API requests, and estimated cost in USD, along with a warning if your average document length approaches the selected model's context limit.