Voyage AI • Image & Text Embeddings
Voyage Multimodal 3 API Pricing, Multimodal RAG & Rate Card (2026)
Complete 2026 developer pricing guide for Voyage Multimodal 3. Benchmark token economics, context window limits, prompt caching discounts, and live inference latency across official provider endpoints.
⚡ 60-Second Multimodal Verdict
✓ Verified October 2026
Unified Text & Image Vector Space • 16k Context
Voyage Multimodal 3 Pricing Verdict: Voyage Multimodal 3 embeds interleaved images, tables, PDFs, and text into a unified vector space. Priced at $0.12 per 1M text tokens and $0.0006 per embedded image, it eliminates the need for separate OCR pipelines by directly encoding visual documents with state-of-the-art precision for enterprise document understanding. Explore models in our Models Directory and compare vision tools in Image & Video Gen Hub.
Text Token Pricing
$0.12 / 1M Tokens
Image Unit Cost
$0.0006 / Image
Context Window
16,000 Tokens (Interleaved)
Primary Benefit
Zero-OCR Visual Document RAG
Architectural Breakdown & Cost Optimization
Voyage Multimodal 3 represents a compelling value proposition in the 2026 developer ecosystem, balancing high accuracy with predictable operating margins. When designing agentic workflows or production RAG systems around Voyage Multimodal 3, prompt engineering and caching strategy represent the single largest levers on your monthly bill.
To maximize unit economics:
- Leverage Prefix Caching: Structure your system prompts and tool definitions at the beginning of the context payload to guarantee cache hits across multi-turn interactions.
- Implement Batch Processing: For non-realtime operations such as data evaluation, document parsing, and backfill tasks, use asynchronous Batch APIs to immediately slash input and output bills in half.
- Monitor Token Bloat: Uncontrolled conversational turns degrade both latency and token economics. Prune older history or summarize intermediate thoughts before context exceeds 32k tokens.
Frequently Asked Questions
What is the token pricing for Voyage Multimodal 3 in 2026?
Voyage Multimodal 3 is billed at $0.18 per million input tokens and $0.00 per million output tokens under standard API tiers.
What is the context window capacity of Voyage Multimodal 3?
Voyage Multimodal 3 supports up to 32,000 tokens, allowing long-document ingestion and extensive conversational history.
Does Voyage Multimodal 3 support prompt caching or batch discounts?
Yes, prompt caching reduces recurring prefix token costs to roughly $0.00 per million tokens for cached blocks.
Voyage Multimodal-3 Architecture: Cross-Modal Vector Space & Joint TCO
Voyage AI's voyage-multimodal-3 projects text passages and images into a single unified 1,024-dimensional shared vector space. In traditional search architectures, engineering teams maintain separate text search indexes and CLIP image embedding models, creating severe modality alignment mismatch. Voyage Multimodal-3 allows users to search visual product catalogs, PDF slides, and technical schematics using natural text queries with native cross-modal similarity matching at $0.12 per 1M text tokens and $0.0008 per image.
| Input Modality |
Pricing Unit |
Vector Dimensions |
Storage / 100k Items |
Cross-Modal Recall |
| Text Documents |
$0.12 / 1M tokens |
1,024 dims |
409.6 MB RAM |
Zero modality gap |
| Images (PNG/JPEG) |
$0.0008 / image |
1,024 dims |
409.6 MB RAM |
Unified vector space |
| Dual Index (CLIP + Text-3) |
Separate API fees |
512 dims + 1536 dims |
819.2 MB RAM (2× storage) |
Requires late-fusion rank merge |
Production Pipeline: Cross-Modal Image-Text Embedding in Python
Embedding images alongside text passages into a single vector database index using Voyage Multimodal:
import voyageai
from PIL import Image
vo = voyageai.Client()
# Embed image file
image = Image.open("architecture_diagram.png")
image_result = vo.multimodal_embed(
inputs=[image],
model="voyage-multimodal-3"
)
image_vector = image_result.embeddings[0]
# Embed text search query to retrieve matching diagrams
query_result = vo.embed(
texts=["Find cloud microservice architecture diagrams with Redis"],
model="voyage-multimodal-3",
input_type="query"
)
query_vector = query_result.embeddings[0]
print(f"Both vectors share dimension {len(query_vector)}")
Production Best Practices & Vision Guidelines
- Single Qdrant/Pinecone Index: Store both image and text vectors in the exact same index namespace. Text queries compute cosine similarity against image vectors with zero translation layers.
- Image Resolution Downsampling: Pre-scale input images to a maximum dimension of 1024px before uploading to eliminate payload transmission latency without degrading embedding fidelity.
- PDF Slide Processing: Extract presentation slides as individual PNG images and pass them directly to `voyage-multimodal-3` to capture visual infographics, charts, and slide headers.
Cross-Modal Alignment Math & Unified Text-Image Search Architecture
Voyage Multimodal 3 maps interleaved sequences of text, screenshots, product photographs, and architectural schematics into a unified 1024-dimensional semantic hyperspace. In traditional dual-encoder architectures (such as CLIP), separate image and text encoders frequently suffer from modality gap distortion where image vectors cluster in distinct subspaces disjoint from text vectors. Voyage Multimodal 3 trains on unified contrastive loss with shared projection heads, ensuring that natural language queries map with high cosine similarity directly to relevant visual image patches.
Multimodal Ingestion Costs and Sizing Matrix
| Content Type |
Equivalent Token Count |
Cost / 1,000 Embeddings |
Vector Database Size (1M Items) |
Retrieval NDCG@10 |
| Pure Text Paragraph |
250 tokens |
$0.030 |
4.1 GB |
0.742 |
| Product Image (1024x1024) |
~1,600 tokens |
$0.192 |
4.1 GB |
0.816 |
| Interleaved Doc + Diagram |
~2,400 tokens |
$0.288 |
4.1 GB |
0.884 |
Interleaved Multimodal Ingestion in Python
Embed structured documents combining text and base64-encoded visual diagrams:
import voyageai
from PIL import Image
vo = voyageai.Client()
img = Image.open("database_architecture.png")
# Embed visual schematic and descriptive text into identical coordinate space
res = vo.multimodal_embed(
inputs=[[img, "Distributed multi-region PostgreSQL cluster with raft consensus"]],
model="voyage-multimodal-3"
)
print("Multimodal vector dimension:", len(res.embeddings[0]))