Embed v4
cohere/embed-v4Multimodal enterprise embeddings: text and images in one vector space, 128K inputs.
Model overview
Context window
128K
tokens
Input price
—
quoted on request
Output price
—
quoted on request
Weekly volume
—
tokens / week
Open weights
No
API only
Capabilities
2
of 11 tags
Description
Embed v4 is an embeddings model from Cohere. Webparam does not stock this model yet and can source it on request — tell us what you need it for and we will come back with availability and a price. Cohere’s stated focus is enterprise retrieval and language stack. The weights are not published, so it is API-only.
Capabilities
- Batch
- Accepts offline batch jobs, which are processed asynchronously at a reduced rate.
- Vision input
- Accepts images alongside text in the prompt.
Strengths & weaknesses
Strengths
- Text + image in one space
- 128K input context
- Enterprise-grade multilingual recall
Weaknesses
- Higher price than Qwen3 Embedding
Pricing
| Rate | Price | Unit |
|---|---|---|
| Input | — quoted on request | — |
| Output | — quoted on request | — |
Batch jobs are supported and settle at a reduced rate against the same unit. This model is not provisioned yet; the rate is quoted on request.
Context window
128Ktokens
covers prompt and response together, so a long input leaves less room for the answer.
Supported features
| Feature | Support |
|---|---|
| Tool calling | Not supported |
| JSON mode | Not supported |
| Streaming | Not supported |
| Vision input | Supported |
| Audio input | Not supported |
| Long context | Not supported |
| Open weights | Not supported |
| Reasoning | Not supported |
| Fine-tunable | Not supported |
| Batch | Supported |
| Caching | Not supported |
Example use cases
Multimodal search
Enterprise RAG
Catalogue matching
Related models
Comparison
Compare against anything →| Attribute | Embed v4 | Qwen3 Embedding | Gemini Embedding |
|---|---|---|---|
| Context window | 128K | 32K | 8K |
| Input price | R0.00 | R0.00 | R0.00 |
| Output price | — not charged | — not charged | — not charged |
| Tool calling | Not supported | Not supported | Not supported |
| JSON mode | Not supported | Not supported | Not supported |
| Vision input | Supported | Not supported | Not supported |
| Open weights | Not supported | Not supported | Not supported |
| Prompt caching | Not supported | Not supported | Not supported |
Best value in each row is highlighted. Blank cells are facts the supplier does not publish, not zeroes.
Provider information
View Cohere →Cohere · Toronto, Canada · est. 2019
Cohere never chased consumer chat. It built the enterprise retrieval stack instead.
Recent releases
Documentation
Frequently asked questions
Input is billed at R0.00 per 1M tokens, and there is no separate output charge. Batch jobs settle at a reduced rate against the same unit. Every figure on this page is an illustrative placeholder.
No. Embed v4 is served through its provider's API only; the weights are not published, so self-hosting is not an option.
Prompt and response share one window: Embed v4 accepts up to 128K tokens in total, so a long input leaves less room for the answer. The median across embeddings models is 32K. Recall is strongest well inside the limit. Test at your real prompt depth before relying on the ceiling.
Qwen3 Embedding is the closest neighbour in the catalogue, with 32K of context at R0.00 per 1M tokens in. Put both in a comparison before wiring a fallback chain, so the trade-off is a decision rather than a default.