Voyage AI

Provides state-of-the-art embedding models and rerankers optimized for retrieval accuracy, used by Anthropic and trusted for RAG pipelines.

Evaluated Mar 06, 2026 (0d ago) vcurrent
Homepage ↗ AI & Machine Learning embeddings vectors retrieval rag semantic-search
⚙ Agent Friendliness
64
/ 100
Can an agent use this?
🔒 Security
81
/ 100
Is it safe for agents?
⚡ Reliability
81
/ 100
Does it work consistently?

Score Breakdown

⚙ Agent Friendliness

MCP Quality
--
Documentation
87
Error Messages
82
Auth Simplicity
92
Rate Limits
78

🔒 Security

TLS Enforcement
100
Auth Strength
80
Scope Granularity
60
Dep. Hygiene
83
Secret Handling
83

TLS enforced on all endpoints; no scope-based key permissions; API key rotation supported via dashboard; data not used for model training per ToS.

⚡ Reliability

Uptime/SLA
83
Version Stability
80
Breaking Changes
78
Error Recovery
83
AF Security Reliability

Best When

You need maximum retrieval accuracy for RAG or semantic search and are willing to pay a premium over commodity embedding APIs.

Avoid When

Your workload is cost-sensitive at scale and OpenAI or Cohere embeddings already meet your accuracy bar.

Use Cases

  • Generating high-quality embeddings for RAG pipelines over large document corpora
  • Reranking retrieved candidates to improve answer relevance before sending to LLM
  • Semantic search over code repositories using voyage-code-3 model
  • Domain-specific embedding with finance or law optimized models
  • Building long-context retrieval pipelines with up to 32K token input support

Not For

  • Generating text completions or chat responses — embeddings only
  • Real-time streaming inference requiring sub-10ms latency
  • On-premise deployment where data cannot leave your infrastructure

Interface

REST API
Yes
GraphQL
No
gRPC
No
MCP Server
No
SDK
Yes
Webhooks
No

Authentication

Methods: api_key
OAuth: No Scopes: No

Single API key passed as Bearer token in Authorization header; keys are per-account with no scope granularity.

Pricing

Model: usage_based
Free tier: Yes
Requires CC: No

Free tier credited at account creation; no ongoing free quota after initial allocation is consumed.

Agent Metadata

Pagination
none
Idempotent
Full
Retry Guidance
Documented

Known Gotchas

  • Input token limit per request is 128K tokens total across all texts in a batch — splitting large batches is required
  • Model names are versioned (e.g. voyage-3) but silent updates can shift embedding space, breaking cached index compatibility
  • Reranker and embedder are separate endpoints with separate pricing — easy to conflate in cost estimation
  • voyage-code-3 is recommended for mixed code+prose but not for pure natural language where voyage-3 outperforms it
  • The Python SDK truncates inputs silently by default — set truncation=False to surface length errors explicitly

Alternatives

Full Evaluation Report

Detailed scoring breakdown, competitive positioning, security analysis, and improvement recommendations for Voyage AI.

$99

Scores are editorial opinions as of 2026-03-06.

5178
Packages Evaluated
26151
Need Evaluation
173
Need Re-evaluation
Community Powered