Updated: October 25, 2026
| Name | Voyage AI |
|---|---|
| Overview | Voyage AI, now part of MongoDB, provides embedding models and rerankers for search and retrieval. These components help select relevant context for AI applications and can reduce retrieval errors, but do not guarantee factual answers from a downstream language model. |
| Key Features & Benefits | – State-of-the-Art Embeddings: Vector-based retrieval with selectable embedding dimensions on supported models; storage and retrieval quality depend on configuration. – Advanced Reranking Models: Improves search relevance for LLMs and AI-powered applications. – Long-Context Handling: Supported models accept up to 32,000 tokens; the current voyage-law-2 limit is 16,000. Check the chosen model rather than applying one limit to every endpoint. – High Efficiency & Low Latency: Models offer different latency, cost and retrieval-quality trade-offs; benchmark performance is not a guarantee for every workload. – Domain & Company-Specific Models: Domain models cover code, finance and legal retrieval. Custom fine-tuning or deployment is a separate sales discussion, not included in every API call. – Seamless API Integration: API and Python-client access supports integration with AI systems, vector databases and LLM frameworks; implementation and evaluation are still required. |
| Use Cases & Applications | – Enhancing Retrieval-Augmented Generation (RAG) for AI models – Improving semantic search and knowledge discovery – Powering chatbots and virtual assistants with better context-awareness – Optimizing legal, finance, and healthcare document retrieval – Boosting code search and recommendation engines – Strengthening AI-driven customer support and e-commerce personalization |
| Who Uses? | – AI engineers and data scientists – Enterprises optimizing search and AI applications – Research institutions and universities – Developers building AI-powered products – Businesses leveraging LLMs for enhanced decision-making |
| Pricing |
Usage-based API pricing in US dollars, not a per-seat subscription. The current pricing table lists voyage-4-lite at $0.02, voyage-4 at $0.06 and voyage-4-large at $0.12 per million input tokens, each with an initial 200 million-token allowance per account. Rerank-3-lite is $0.02 and rerank-3 $0.05 per million processed tokens; the table lists 200 million free tokens for each. Reranking counts the query repeatedly for the documents being ranked, not just once per request. Other models and multimodal inputs have their own units and allowances. Batch API usage has a separate discounted rate and does not use free-token balances. Some introductory text names older models differently from the current tables; verify the exact model and account allowance before a large job. Custom models/deployment are quoted separately. |
| Tags | AI Search, AI Retrieval, RAG, Embeddings, Rerankers, NLP, Vector Search, LLM Optimization |
| App Available? | Developer API and Python client with a browser dashboard, rather than a standalone consumer chat app. Available cloud and custom deployments depend on the offer; no separate native mobile app was verified. |
Practical use: For a document-search application, embed the documents and queries, retrieve candidates, then optionally rerank them before sending context to an LLM. Check relevance and final answers against source documents. Legal, financial and healthcare document retrieval is not authorization for autonomous professional decisions. Review the data-use and opt-out terms before uploading confidential information; opting out applies prospectively and may affect free usage.
Sources: Product overview; Embedding models; API pricing; MongoDB acquisition announcement; Terms of service.
Image context: The existing Voyage AI homepage screenshot shows an earlier green-and-white search and retrieval landing page. Its capture date is unknown; it identifies the provider rather than demonstrating a tested search result or current benchmark performance.
