OpenAI vs Voyage AI embeddings: which one for your RAG?

OpenAI's text-embedding-3 models and Voyage AI's Voyage 4 family compared on what matters for retrieval: quality on your documents, price, vector size, query and document modes, and switching cost. With a head-to-head benchmark.

  • embeddings
  • rag
  • openai
  • voyage-ai
  • model-comparison

Download the benchmark report (PDF)

OpenAI's embedding models are the default in most RAG tutorials and frameworks. Voyage AI builds embedding models focused on retrieval, and teams often consider it as the alternative. If you're choosing between them, or thinking of switching, the choice deserves more than a glance at a leaderboard.

This post compares the two families on the points that matter in production. The last section is a head-to-head benchmark of five models: OpenAI text-embedding-3-small and 3-large against Voyage 4 Lite, Voyage 4 and Voyage 4 Large.

The line-ups

OpenAI Voyage AI
Models text-embedding-3-small, text-embedding-3-large Voyage 4 Lite, Voyage 4, Voyage 4 Large
List price per 1M tokens $0.02 (small), $0.13 (large) $0.02 (Lite), $0.06 (4), $0.12 (Large)
Dimensions (as run in Truvec) 1,536 (small), 3,072 (large) 1,024
Query vs document mode Same embedding for both Separate query and document input types

Prices are list prices at the time of writing; check the providers' pricing pages.

Differences that matter

Price tiers. Both have a cheap model at the same list price. Voyage adds a middle tier, which matters when the cheap model isn't good enough and the large one costs more than you want to pay.

Vector size. OpenAI 3-large produces 3,072 dimensions, three times Voyage's 1,024. At a million chunks that's about 12 GB of raw vectors against 4 GB, which matters if your vector database keeps them in memory. OpenAI's v3 models can return shortened vectors (the dimensions parameter), trading some quality for size; test the shortened size if you plan to use it.

Query and document modes. Voyage embeds questions and passages differently when told which is which. This asymmetry suits retrieval, where short questions are matched against long passages. OpenAI uses one embedding for both, which is simpler but gives the model less to work with.

Ecosystem. OpenAI is supported everywhere by default. Voyage is supported by the major frameworks and vector databases too, but if you already run everything on OpenAI, adding a second provider means a second key, a second bill and a second set of rate limits.

Switching cost. Vectors from different models aren't compatible. Switching means re-embedding the whole corpus and re-indexing it. That cost is a reason to compare carefully before you commit, not a reason to never switch.

How to decide

  1. Run both families on your own documents, at the top k you use in production.
  2. If the results are within the noise (a couple of points with 100 questions), choose on price, vector size and operations. Staying on the provider you already use is a valid tie-breaker.
  3. If one family is clearly ahead, check that the lead holds across question types, using the example questions where the models disagree.
  4. Compare the mid-tier models too. "Large vs large" isn't the only fair fight: a cheaper model that ties with an expensive one is the better buy.

The benchmark below compares the five models on developer documentation and support articles.

Real benchmark - generated by Truvec

Benchmark: OpenAI vs Voyage AI on a technical knowledge base

Corpus
Docker documentation: manuals, FAQs, billing and support pages
Documents
261
Chunks
1,944
Chunking
recursive, 1000 chars, 200 overlap
Golden set
100 questions, human-reviewed
Models
5 compared
Top k
5
Run
2026-11-14

Download the full report (PDF) Every question, with the passages each model retrieved.

Key results

  • Voyage 4 Large ranks first, with an MRR of 0.922 and a hit rate of 99.0% in the top 5.
  • One model is within 0.02 MRR of the winner (Voyage 4): on this corpus they are practically tied.
  • Hit rate ranges from 95.0% to 100.0% across the 5 models.
  • Cheapest model within 2 points of the winner's hit rate: Voyage 4 Lite, $0.0067 to embed this corpus against $0.04.
  • Fastest to embed a query: OpenAI text-embedding-3-small (4.8 ms per query, measured in batch from our servers).

Retrieval quality

Retrieval quality at top 5, ranked by MRR, then recall@k
ModelHit rate@5Recall@5MRRnDCG@5Precision@5
1Voyage 4 Largevoyage - 1024 dims99.0%97.5%0.9220.92420.6%
2Voyage 4voyage - 1024 dims100.0%99.5%0.9150.92921.2%
3Voyage 4 Litevoyage - 1024 dims99.0%97.5%0.8930.90720.6%
4OpenAI text-embedding-3-largeopenai - 3072 dims99.0%97.5%0.8670.88520.6%
5OpenAI text-embedding-3-smallopenai - 1536 dims95.0%93.5%0.7690.80719.8%

Hit rate by position

How many questions the first results already answer. A model that finds the answer at position 1 lets you send fewer chunks to the LLM.

Hit rate by position: share of questions answered by the first j results
ModelTop 1Top 2Top 3Top 4Top 5
1Voyage 4 Largevoyage - 1024 dims87.0%94.0%99.0%99.0%99.0%
2Voyage 4voyage - 1024 dims85.0%95.0%99.0%99.0%100.0%
3Voyage 4 Litevoyage - 1024 dims81.0%95.0%99.0%99.0%99.0%
4OpenAI text-embedding-3-largeopenai - 3072 dims79.0%88.0%94.0%98.0%99.0%
5OpenAI text-embedding-3-smallopenai - 1536 dims63.0%84.0%92.0%94.0%95.0%

Cost, speed and size

Embedding prices are the providers' list prices. Corpus cost is a one-time cost per model; the vector size is raw float32 storage, before the vector database's own overhead.

Cost, speed and size on this corpus
ModelPrice / 1M tokensEmbed the corpus1,000 queriesQuery embeddingSearch p95Vectors
1Voyage 4 Largevoyage - 1024 dims$0.12$0.04$0.002418 ms61 ms7.6 MB
2Voyage 4voyage - 1024 dims$0.06$0.02$0.001244 ms56 ms7.6 MB
3Voyage 4 Litevoyage - 1024 dims$0.02$0.0067$0.000415 ms25 ms7.6 MB
4OpenAI text-embedding-3-largeopenai - 3072 dims$0.13$0.04$0.002613 ms124 ms22.8 MB
5OpenAI text-embedding-3-smallopenai - 1536 dims$0.02$0.0067$0.00044.8 ms32 ms11.4 MB

Score separation

Average similarity of the best relevant chunk and of the best irrelevant one. A wide gap makes a similarity threshold reliable; a narrow one means right and wrong chunks look alike to the model.

Score separation: how clearly each model scores the right chunk above the best wrong one
ModelRelevant chunkBest distractorGap
1Voyage 4 Largevoyage - 1024 dims0.6530.6070.046
2Voyage 4voyage - 1024 dims0.6920.6460.046
3Voyage 4 Litevoyage - 1024 dims0.6650.6220.044
4OpenAI text-embedding-3-largeopenai - 3072 dims0.6870.6500.037
5OpenAI text-embedding-3-smallopenai - 1536 dims0.7150.6870.027

Example questions

Rank of the first relevant chunk for each model; miss means it wasn't in the top 5. These are the questions where the models disagree most.

Example questions from the golden set
QuestionVoyage 4 LargeVoyage 4Voyage 4 LiteOpenAI text-embedding-3-largeOpenAI text-embedding-3-small
When both JIT and SCIM are enabled, how are organization roles and account deactivation handled?
Expected passage
Roles don't move back and forth. JIT sets the organization role only when it first adds the user. A later SCIM update can change that role, and the next sign-in leaves the SCIM role in place. When a user isn't assigned to the Docker application in the IdP, the next synchronizatio
1125miss
When multiple Compose files are supplied with `-f`, how are their service settings combined, and which file takes precedence?
Expected passage
Compose merges files in the order they're specified on the command line. Subsequent files may merge, override, or add to their predecessors. For example: ```console $ docker compose -f compose.yaml -f compose.admin.yaml run backup_db ``` The `compose.yaml` file might specify a `w
12314
What command builds the Docker image for the Spring Boot hello world app, and how can I check the size of the resulting image?
Expected passage
```dockerfile CMD ["./mvnw", "spring-boot:run"] ``` And with that, you should have the following Dockerfile: ```dockerfile FROM eclipse-temurin:21.0.8_9-jdk-jammy WORKDIR /app COPY .mvn/ .mvn COPY mvnw pom.xml ./ RUN ./mvnw dependency:go-offline COPY src ./src CMD ["./mvnw", "spr
23215
How can I combine multiple Compose files, and what issue might arise as the configuration becomes more complex?
Expected passage
# Use multiple Compose files This section contains information on the ways you can work with multiple Compose files. Using multiple Compose files lets you customize a Compose application for different environments or workflows. This is useful for large applications that may use d
23211
How do I push an image to Docker Hub, and what tagging restriction applies?
Expected passage
## Push an image to Docker Hub Select the image from the list, select the **More options** button and select **Push to Hub**. > [!NOTE] > > You can only push an image to Docker Hub if the image belongs to your Docker ID or your organization. That is, the image must contain the co
miss5missmiss1
Which organizational roles and subscription plans are needed to create organization access tokens in Docker?
Expected passage
> [!WARNING] > > OATs don't work with Docker Desktop or Image Access Management. For > those features, use > PATs instead. ## Best practices - Set an expiration date and rotate tokens on a schedule - Grant only the repositories and scopes each job needs - Review the **Last used**
1114miss
Which Docker Desktop versions are supported under the Pro, Team, and Business plans?
Expected passage
### Supported versions Docker Desktop support is scoped by plan: | Plan | Supported versions | Machines per subscription | | :--- | :--- | :--- | | Docker Pro | Latest only | One | | Docker Team | Latest only | One per subscription seat | | Docker Business | Latest plus versions
11123
What tasks can Docker Compose perform for a multi-container application besides starting its services?
Expected passage
# Docker Compose Docker Compose is a tool for defining and running multi-container applications. It is the key to unlocking a streamlined and efficient development and deployment experience. Compose simplifies the control of your entire application stack, making it easy to manage
22311

Notes

  • The corpus is a subset of the public Docker documentation (Apache 2.0), taken from github.com/docker/docs at commit 553482c: the Desktop, Docker Hub, Compose, accounts, security and subscription and billing manuals, the FAQs, the support pages and the getting-started guides. Release notes and pages shorter than 600 characters were left out, and pages were cut at 40,000 characters.
  • Several posts on this blog use this corpus and the same 100 questions, each with its own selection of models. A model can differ by a question between two of these runs: questions are embedded again at each run, and vector search is approximate.
  • With 100 questions, a gap of a couple of points between two models is within the noise.

Methodology

Truvec split the documents into chunks, generated a golden set of questions with gpt-5.6-luna (each question is linked to the chunks that answer it), embedded the corpus and the questions with every model, and retrieved the top 5 chunks for each question by cosine similarity. Every model saw the same chunks and the same questions. Scores are averaged over all 100 questions.

Hit rate
Share of questions with at least one relevant chunk in the top k results. The most direct answer to "does the right passage reach the LLM?".
Recall
Share of the relevant chunks found in the top k, averaged over questions. Matters when an answer spans several chunks.
MRR
Mean reciprocal rank: 1 when the first relevant chunk is ranked first, 1/2 when second, 0 when missing. Rewards putting the answer at the top.
nDCG
Normalized discounted cumulative gain: how close the ranking of all relevant chunks is to the ideal ranking.
Precision
Share of the top k results that are relevant. Low precision means more noise sent to the LLM.

Downloads: full report (PDF) - benchmark.json - benchmark.csv - this post as Markdown

Voyage 4 Large won here, on these documents. Find the model that wins on yours: run the same benchmark on your documents with Truvec.