Voyage AI
Get high-accuracy embedding and reranking models for semantic search and RAG, now reachable directly inside MongoDB Atlas after MongoDB's acquisition, still available standalone too.
The verdict on Voyage AI: Teams whose RAG system's performance bottleneck is actually retrieval quality, not the generation model Voyage AI's thesis is that retrieval quality is the actual bottleneck in most RAG systems, not the generation model. Pricing: From $0.02/M tokens (voyage-4-lite).
Best For
Teams whose RAG system's performance bottleneck is actually retrieval quality, not the generation model
Standout Feature
Purpose-built embedding and reranking models that outperform general-purpose embeddings on retrieval specifically
TL;DR
A real quality upgrade for RAG-heavy applications, it's embeddings-only, not a general AI platform.
Alternatives
Fit Score: use-case reference
How this is scoredVoyage AI scored 1 to 5 on five dimensions for the named use cases below. Part of our catalog-wide use-case reference, not a product-quality review rating.
A team whose RAG system's actual performance bottleneck is retrieval quality rather than the generation model
4.2/5- Capability
- 5/5
- Reliability
- 4/5
- Value
- 4/5
- Ease
- 4/5
- Compatibility
- 4/5
Voyage AI's purpose-built embedding and reranking models are designed specifically to outperform general-purpose embeddings on retrieval, which the tool's own take frames as worth evaluating once you have evidence retrieval failures are degrading RAG output quality.
A team looking for a general-purpose AI platform rather than a narrow embeddings and reranking service
2.2/5- Capability
- 1/5
- Reliability
- 3/5
- Value
- 2/5
- Ease
- 3/5
- Compatibility
- 2/5
Cons state the focus is embeddings-only, and the tool's own take is explicit that the scope is narrow: this is an embeddings and reranking service, not a general AI platform.
Editorial judgment grounded in this tool's own published pricing, features, and our existing pros and cons, not a certification or a measured benchmark. Product quality is assessed separately under our review methodology.
Overview
Not yet reviewedProduct quality is assessed from original evidence. How reviews work.
Voyage AI is a specialized AI company (now part of MongoDB) focused exclusively on embedding models and reranking models for retrieval-augmented generation and semantic search applications, with the thesis that retrieval quality is the primary bottleneck in most RAG system performance and that purpose-built retrieval models outperform general-purpose embeddings. Voyage's embedding models are calibrated specifically for the retrieval task, optimized for the semantic similarity judgment that determines whether a retrieved document is relevant to a query, rather than the broader text understanding objectives that general language model embeddings optimize for. The company publishes domain-specific models for legal, finance, code, and multilingual retrieval where the vocabulary and semantic relationships of each domain produce meaningfully different embedding spaces than general text. Rerank models take a query and a set of retrieved documents and reorder them by relevance, providing a second-stage quality filter that improves the final retrieval set beyond what first-stage vector search achieves.
Voyage's reranking approach can improve RAG answer accuracy by 5–15% on enterprise benchmarks compared to embedding-only retrieval. The API is compatible with the OpenAI embedding API format for easy integration. Free tier provides 50 million tokens per month. Voyage is most commonly deployed as the embedding and reranking layer in enterprise RAG systems where retrieval precision has direct impact on the accuracy of the AI output.
The source of the legacy catalog rating and review count has not been substantiated. Neither is presented here as a verified review rating or included in review structured data.
Workflow Fit Scores are separate, use-case-specific editorial judgments, not user-review averages. Read the Fit Score methodology.
- • Excellent retrieval quality
- • Strong reranking
- • Easy integration
- • Paid for production volume
- • Embeddings-only focus
Key Features
- Top-ranked embeddings
- Reranking models
- Domain-specific variants
- Simple API
Trust & Data
Verified 2026-09
- Training on your data
- Opt-out availableVoyage AI's own privacy policy, still hosted on voyageai.com rather than migrated to mongodb.com even after the 2025 MongoDB acquisition, states that by default, unless you opt out, you grant Voyage AI a perpetual license to use customer content submitted via the API to train, improve, and further develop its AI models; opting out stops this only prospectively, since data submitted before opting out remains subject to the training license. This directly contradicts a separate MongoDB documentation page for the MongoDB-integrated Voyage AI API, which states customer data from that specific feature is not used to train models at this time, so which policy applies depends on which access route a customer uses.
- Compliance
- Not published
- Retention
- No specific general retention period is stated beyond the training opt-out clause, under which customer content is deleted immediately after processing once a customer opts out, but only prospectively.
- Export
- Partial export
Sources: voyageai.com, mongodb.com, blog.voyageai.com
As published by the vendor. Verify independently before purchase decisions.
Our Take
The purpose-built embedding and reranking models are designed specifically for that problem, and they outperform general-purpose embeddings on retrieval tasks. Now part of MongoDB, the models are directly accessible inside MongoDB Atlas, but still available standalone. The honest scope is narrow: this is an embeddings and reranking service, not a general AI platform. Worth evaluating when you have evidence that retrieval failures are what's degrading your RAG system's output quality.