Voyage AI Free Tokens: 200M Free Embedding Tokens (2026)

Voyage AI gives every account a one-off allowance of free tokens on its current embedding and reranking models: 200 million tokens on the voyage-4 family, voyage-code-4, rerank-3 and rerank-3-lite, plus 200M text tokens and 150 billion pixels on its multimodal models. Smaller domain models (finance, law, code-2) get 50 million. You do not apply for anything. You sign up, create an API key, and free tokens are used before any paid usage. The catch is that older models and the Batch API get no free tokens, and the higher rate limits require a payment method on file.

What you get, model by model

Voyage lists free allowances per model in the pricing tables on its documentation site. The table below follows those tables as of October 2026. Prices are what you pay per million tokens once the free allowance for that model is used.

Everything else is in Voyage's older models table, which the pricing page says gets no free tokens. That includes voyage-3.5 ($0.06), voyage-3.5-lite ($0.02), voyage-3-large, voyage-context-3 and voyage-code-3 ($0.18 each), voyage-multilingual-2 ($0.12), the voyage-01/02 series ($0.10), and rerank-1, rerank-2 and rerank-2.5 ($0.05) plus their lite versions ($0.02).

The pricing page contradicts itself. Its introductory text still says voyage-code-3, voyage-multilingual-2, rerank-2.5, rerank-2.5-lite, rerank-2 and rerank-2-lite are free, but the tables on the same page put them under older models with no free tokens. If you want to be sure your usage is free, build on the models listed as free in the tables: voyage-4, voyage-code-4 and rerank-3.

How multimodal usage is counted

Multimodal requests are charged on text tokens and on image pixels. Voyage treats every video frame as an image. Images under 50,000 pixels are charged as 50,000 pixels, and images over 2 million pixels are downsampled and charged as 2 million. So once the free allowance is used, a single image costs between $0.00003 and $0.0012. Voyage's own example: one input with 1,000 text tokens and two 4-megapixel images costs $0.00012 + 2 × $0.0012 = $0.00252.

The free 150 billion pixels go a long way. At the 2-million-pixel cap per image, 150B pixels ÷ 2M = 75,000 large images before you pay for pixels. At the 50,000-pixel minimum, it is 3 million small images.

How reranker usage is counted

Reranking is billed on processed tokens, which Voyage defines as (query tokens × number of documents) + the sum of all document tokens. A long query multiplies quickly. Voyage's price estimate assumes 100 documents where each query-plus-document pair is 500 tokens, which works out to about $0.0025 per rerank-3 request and $0.001 per rerank-3-lite request once free tokens are gone. On that same assumption (about 50,000 processed tokens per request), the 200M free tokens cover roughly 4,000 rerank requests.

Who qualifies

Voyage's pricing page says the free tokens are available for every account. It publishes no other conditions: no student or startup check, no country list on the pricing page, and no requirement to use MongoDB. You need a Voyage account at dash.voyageai.com and an API key.

• New and existing accounts: the wording is per account, not per new signup. New models released later (for example voyage-code-4 in August 2026 and rerank-3 in September 2026) came with their own 200M free tokens, so existing accounts could use them.
• Organizations: billing and rate limits are managed at the organization level in the dashboard. Projects inside an organization share the organization's rate limits unless an Admin sets lower project limits.
• Payment method: not needed to sign up. Adding one moves you to usage Tier 1, which unlocks the published rate limits.

If you are building on MongoDB Atlas, Voyage models are also available through Atlas. The voyage-code-4 announcement says the first 200M tokens are free through both the Voyage API and MongoDB Atlas. This page covers the Voyage API route.

How to start using the free tokens

1. Create an account at dash.voyageai.com.
2. In the dashboard, create an API key. Keep it server-side; anyone with the key spends your allowance.
3. Install the official Python client (`pip install voyageai`) or call the REST API directly.
4. Pick a model that has free tokens, for example voyage-4 for documents or rerank-3 for reranking.
5. Send a request. Free tokens are applied automatically; there is no coupon or activation step.
6. Add a payment method under the Organization section if you need Tier 1 rate limits or plan to go past the free allowance.

A minimal embedding call in Python looks like this:

import voyageai

vo = voyageai.Client()  # reads VOYAGE_API_KEY from the environment
result = vo.embed(
    ["How do I reset my password?", "Refund policy for annual plans"],
    model="voyage-4",
    input_type="document",
)
print(len(result.embeddings), result.total_tokens)

The `total_tokens` value in the response tells you how many tokens the call used, which is the easiest way to estimate how far the free allowance will stretch for your data.

Rate limits and usage tiers

Voyage's rate-limit page publishes limits for usage Tier 1, which you reach by adding a payment method. Limits rise automatically as you spend: Tier 2 (at least $100 paid) doubles them, and Tier 3 (at least $1,000 paid) triples them. Voyage says tier qualification is based on billed usage and excludes free tokens for current models, and that you are never downgraded once you qualify.

Voyage does not publish rate limits for accounts that have not added a payment method. If you only want to test with free tokens, expect lower limits until you add one, and check the Rate Limits section of your organization in the dashboard for the exact numbers on your account.

What it costs after the free tokens

Once a model's free allowance is used, you pay per token at the prices above, billed monthly. Voyage's FAQ says it charges the card on file around the 2nd of each month for the previous month's usage. You can also prepay: prepaid credits are applied to your monthly invoice, expire after one year, and are non-refundable.

Worked examples

• Embedding a 10,000-page knowledge base with voyage-4: at about 600 tokens per page that is 6M tokens, comfortably inside the 200M free allowance. Re-embedding it 30 times would still use only 180M.
• A large corpus with voyage-4-large: 500M tokens would use the 200M free tokens, then 300M paid tokens × $0.12 per million = $36.
• The same 500M tokens on voyage-4-lite: 300M paid tokens × $0.02 per million = $6.
• Through the Batch API: free tokens do not apply, so all 500M tokens on voyage-4 would be billed at the batch rate, which Voyage describes as a 33% discount on the standard $0.06 per million.

The Files API used for batch jobs charges $0.05 per GB per month for storage, and files are deleted automatically after 30 days.

Batch API: cheaper, but no free tokens

Voyage's Batch API processes large jobs within a 12-hour window at a 33% discount. The pricing page is explicit that free token credits do not apply to Batch API usage: successfully processed batch tokens are billed at the discounted rate and do not reduce your free balance. If you are still inside your free allowance, the standard endpoint is cheaper (it is free). Switch to Batch only for volume beyond the free tokens, or for jobs where you would pay anyway.

Getting the most from 200M tokens

• Pick the model that has free tokens. Calling voyage-3.5 or voyage-code-3 costs money from the first token; voyage-4 and voyage-code-4 do not.
• Use input_type. Voyage's API distinguishes queries from documents. Setting it correctly improves retrieval without costing extra tokens.
• Shrink vectors, not text. The voyage-code-4 announcement describes Matryoshka embeddings at 2048, 1024, 512 and 256 dimensions with int8 and binary quantization. Smaller vectors cut storage costs in your vector database without using more tokens.
• Rerank fewer candidates. Reranker cost scales with query tokens × document count. Retrieving 50 candidates instead of 200 cuts processed tokens by roughly three quarters.
• Chunk once, cache embeddings. Store vectors and only re-embed documents that change. Most free allowances are burned by re-running the same corpus during development.
• Combine with a free LLM tier. Embeddings and reranking are only half of a RAG app. Pair Voyage with a free generation API such as the Gemini API free tier or Mistral's free API mode.

How Voyage compares with other free embedding options

For a wider list of free API keys across providers, see our guide to free AI API keys.

Recent changes to the offer

• February 24, 2025: MongoDB announced it was acquiring Voyage AI. The press release said Voyage's models would remain available through voyageai.com, AWS Marketplace and Azure Marketplace.
• August 13, 2026: Voyage released voyage-code-4 at $0.12 per million tokens, with the first 200M tokens free through the Voyage API and MongoDB Atlas.
• September 30, 2026: Voyage released rerank-3 ($0.05 per million) and rerank-3-lite ($0.02 per million), with the first 200M tokens free. Voyage said rerank-2.5 remains available but recommended moving to rerank-3.
• Current pricing tables: voyage-3.5, voyage-code-3, voyage-multilingual-2 and the rerank-2.5 series now sit in the older-models table with no free tokens. If you built on those models while they were free, check whether your usage is now billed.

Common mistakes

• Assuming every Voyage model is free. Only the models listed with a free allowance are. Older models are billed from the first token.
• Running a big backfill through the Batch API while free tokens remain. Batch usage is always billed and never draws on the free balance.
• Underestimating reranker tokens. Query tokens are multiplied by the number of documents, so a 200-token query against 100 documents is 20,000 tokens before counting the documents themselves.
• Testing without a payment method and hitting limits. Published rate limits start at Tier 1, which needs a payment method.
• Leaving data-training settings at default. Voyage's FAQ says customers can opt out of Voyage storing and using their data for training, but only an organization Admin can change it, and the opt-out cannot be reversed from the dashboard.

Voyage does not publish an expiry date for free tokens. Prepaid credits are different and expire after one year.