Cohere Embed Pricing: embed-v4, embed-v3 & Multilingual Options (June 2026)
Independent pricing reference for Cohere's embedding models. Text and image rates, the 128K context window, multilingual performance data, and enterprise deployment options.
How much does Cohere Embed cost, and is there a free tier?
Cohere embed-v4 costs $0.12 per million tokens for text and $0.47/M for image tokens, with no batch API discount. The legacy embed-v3 generation is $0.10/M. Cohere's free trial tier allows 100 API calls per minute (rate-limited, no lifetime token cap) for development and prototyping.
embed-v4 takes a 128K-token context window and multilingual coverage across 100+ languages. At $0.12/M it is 6x the price of OpenAI text-embedding-3-small, so the premium is justified mainly by non-English retrieval quality.
Current Pricing
| Model | $/M tokens | Batch | Dims | Context | Best for |
|---|---|---|---|---|---|
| embed-v4 (text) | $0.12 | None | 1,536 | 128K tokens | Multilingual text search |
| embed-v4 (images) | $0.47 | None | 1,536 | N/A | Visual search, cross-modal |
| embed-v3 (legacy) | $0.10 | None | 1,024 | 512 tokens | Migrate to v4 |
Multilingual Strength: The Core Case for Cohere
Cohere's primary competitive advantage is multilingual quality. Embed-v4 supports 100+ languages with demonstrably better performance on non-Latin scripts compared to OpenAI text-embedding-3-small. Published benchmarks show 15-20% retrieval quality improvement for Arabic, Hindi, Japanese, and Chinese content.
For applications that operate purely in English, this advantage largely disappears. At $0.12/M tokens (6x more than OpenAI text-embedding-3-small at $0.02/M), Cohere is hard to justify for English-only RAG. The math changes when your retrieval quality requirements in non-English languages are non-negotiable.
Image Embedding with embed-v4
At $0.47/M image tokens, Cohere's multimodal capability enables a shared embedding space for text and images. A text query can retrieve relevant images, and image inputs can retrieve relevant text - useful for e-commerce product search, media libraries, and cross-modal RAG. The shared vector space (configurable Matryoshka dimensions up to 1,536) means you can index images and text into the same vector database collection.
Enterprise Deployment: AWS Marketplace & On-Premises
Cohere has a strong enterprise deployment story. Models are available through:
- -AWS Bedrock: Cohere Embed is available as a Bedrock foundation model. Data stays within your AWS VPC. Bedrock pricing applies (may differ from direct).
- -Azure AI Foundry: Available via Microsoft's model marketplace for enterprise Azure customers.
- -Private deployment: Cohere offers private cloud and on-premises deployment for regulated industries where data residency is mandatory.
When to Pick Cohere vs OpenAI vs Voyage
- - Multilingual corpus (100+ languages)
- - Non-Latin script accuracy matters
- - AWS/Azure enterprise deployment
- - Multimodal (text + image) search
- - English-only RAG applications
- - Budget is primary constraint
- - Existing OpenAI API integration
- - Batch API 50% discount needed
- - Best accuracy-to-price ratio
- - Domain-specific models (code, law)
- - Long-context documents (32k)
- - Mix index/query model sizes
Free Tier
Cohere's free trial tier allows 100 API calls per minute with rate-limited access to all embedding models. There is no lifetime token cap - the limitation is rate, not volume. This is sufficient for building and testing RAG pipelines but not production workloads. For more than 100 calls/minute, you need a paid plan.