Cohere Embed 5 vs. Voyage 4 Large, Gemini Embedding 2, and OpenAI: A 2026 Embedding Model Comparison
Cohere has released Embed 5, a new embedding model family targeting enterprise search, RAG, and agentic retrieval. The release lands in a crowded 2026 retrieval stackβone where Voyage 4 Large, Gemini Embedding 2, and OpenAI's latest embedding models all compete for the same production workloads.
The family ships in two tiers:
- Embed 5 Pro β optimized for maximum retrieval quality
- Embed 5 Fast β optimized for latency and cost on the live query path
Both tiers accept text, images, and fused text-plus-image inputs. Both cover 100+ languages and read up to 128K tokens. The key design choice: Pro and Fast share a single embedding space, so you can index with one and query with the other.
Availability: Deployable Today
Both tiers are generally available on the Cohere API and Model Vault, Microsoft Foundry, and Amazon SageMaker. Private VPC or on-prem serving runs through vLLM.
What Cohere Shipped
The API model IDs are embed-v5.0-pro and embed-v5.0-fast, per Cohere's model docs. Key specifications:
| Spec | Detail |
|---|---|
| Output dimensions | 2048 (default), 1536, 1024, 768, 512, or 256 |
| Embedding formats | float, int8, or binary |
| Pro pricing | $0.12 per 1M text tokens |
| Fast pricing | $0.08 per 1M text tokens |
| Image input pricing | $0.40 per 1M tokens (both tiers) |
Embed 5 can embed a page image directly, or fuse an image with its metadata into a single vector. That matters for scanned pages, slide decks, schematics, and charts, where text extraction drops information.
Pro and Fast: One Index, Two Query Paths
Cohere tested every corpus and query pairing across 40 development datasets. Normalized to Pro-plus-Pro at 100:
- Pro index queried with Fast: 98.4
- All-Fast setup: 96.6
Cohere's recommended pattern is to index with Pro and query with Fast. One constraint applies: both sides must use the same output dimension.
The split targets agentic workloads. An agent may issue dozens of searches per task, so paying Pro-level cost on every query becomes prohibitive. Indexing once with Pro and serving queries through Fast preserves most quality while cutting live-path costs.
How Embed 5 Compares to Voyage 4 Large, Gemini Embedding 2, and OpenAI
The 2026 embedding landscape splits along three axes: multimodal support, dimensional flexibility, and price-per-token on high-volume retrieval.
Voyage 4 Large remains a benchmark leader on text-only retrieval tasks, with strong performance on domain-specific corpora such as legal and financial documents. Its weakness is narrower multimodal support compared to Embed 5's fused text-plus-image vectors.
Gemini Embedding 2 benefits from deep integration with Google's ecosystem and competitive multilingual coverage. Teams already running on Vertex AI gain operational simplicity, though per-token costs on image inputs tend to run higher than Cohere's $0.40 per 1M tokens.
OpenAI's embedding models offer the broadest ecosystem tooling and the easiest integration path for teams already on the OpenAI stack. They remain a default choice for general-purpose RAG, but Embed 5's shared Pro/Fast embedding space is a structural advantage OpenAI does not currently match.
Where Embed 5 differentiates:
- Shared embedding space across tiers β index with Pro, query with Fast, no re-embedding required
- Matryoshka-style dimension flexibility β six output sizes from 2048 down to 256
- Binary and int8 quantization β native support reduces storage and memory costs at scale
- Fused multimodal vectors β a single vector for an image plus its metadata, not separate embeddings
The Bottom Line
For enterprise teams running high-volume agentic retrieval in 2026, Embed 5's Pro-index/Fast-query pattern is the most practical cost-control mechanism available in a production embedding model. Whether it beats Voyage 4 Large or Gemini Embedding 2 depends on your corpus: text-only retrieval benchmarks may still favor Voyage, while multimodal and agentic pipelines favor Cohere's tiered architecture.
Both tiers are live now on Cohere, Microsoft Foundry, and Amazon SageMaker.
via MarkTechPost