Cohere Rerank

Open Embed

Hydrating page from local storage...

Retrieve with embeddings. Rerank with the original text. See exactly which results move.

Embedding connection

Embed endpoint

Model / deployment

Authentication Mode

API Key

Rerank connection

Rerank endpoint

Model / deployment

Uses the Embed connection's credentials. This is allowed only for the same HTTPS origin.

Embedding shape

Output dimensions

Float embeddings · search_document for sources, search_query for queries. Overlong inputs return an error rather than being silently truncated.

Rerank max tokens per document

Rerank truncates longer documents to this token limit.

Advanced

Embed endpoint: https://api.cohere.com/v2/embed for Cohere direct or https://<resource>.services.ai.azure.com/providers/cohere/v2/embed for Foundry. The latest models are Cohere-Embed-V5-Fast and Cohere-Embed-V5-Pro on Foundry, or embed-v5.0-fast and embed-v5.0-pro on Cohere direct. Both support text and images, a 128K-token context, and 256, 512, 768, 1024, 1536, or 2048 dimensions. The existing default remains embed-v4.0 on Cohere direct and embed-v-4-0 on Foundry.

Rerank endpoint: https://api.cohere.com/v2/rerank for Cohere direct or https://<resource>.services.ai.azure.com/providers/cohere/v2/rerank for Foundry. The model / deployment name is rerank-v4.0-pro on Cohere direct and defaults to Cohere-rerank-v4.0-pro on Foundry.

Use the complete endpoint URL and your actual deployment name. A resource base URL or OpenAI /embeddings endpoint does not accept this mixed-input payload. Choose saved credentials from Vault.

Embed 5 launch prices (USD): Fast $0.08 / 1M text tokens, Pro $0.12 / 1M text tokens, and both $0.40 / 1M image tokens. These Cohere-published rates are reference estimates for Foundry deployments.

Foundry Global Standard prices (USD): Embed v4 $0.12 / 1M text tokens and $0.47 / 1M image tokens. Rerank v4 Fast $2.00 / 1K search units; Pro $2.50 / 1K search units. Estimates use response billing metadata; deployment tiers and negotiated charges can differ.

Mixed sample loaded. Configure your connection to create real embeddings.

What are you looking for?

Retrieve candidates

Return top results

The first search creates real Embed vectors for this dataset. The query is embedded, the closest candidates are retrieved, and their text or image captions are sent to Rerank. Raw vectors and images are never sent to Rerank.

Input dataset 8 / 96

One row, one embedding. Labels identify sources; they are not embedded.

Import an exported embedding dataset

Sources + float vectors + model metadata. No credentials. Maximum 40 MB.

Label

Input type Text to embed and rerank

Label

Input type

Replace image

PNG, JPEG, WebP, GIF. 20 MB combined; 16 MP per image.

Caption for Rerank (not embedded)

Label

Input type Text to embed and rerank

Label

Input type

Replace image

PNG, JPEG, WebP, GIF. 20 MB combined; 16 MP per image.

Caption for Rerank (not embedded)

Label

Input type Text to embed and rerank

Replace image

PNG, JPEG, WebP, GIF. 20 MB combined; 16 MP per image.

Label

Input type Text to embed and rerank

Label

Input type Text to embed and rerank

Label

Input type Text to embed and rerank

From candidates to better matches

Import vectors from Embed, or search the sample dataset to create real embeddings first. Compare the retrieved order with Rerank's results here.

Sample descriptions and illustrations are synthetic. All vectors and scores come from your configured models; none are simulated.

Auto-save stores only connection and model settings in this browser, with API keys and client secrets encrypted. Each completed request saves its sources, vectors, queries, rankings, and payloads in browser session history. Profiles contain settings; session ZIPs also contain the working data and never include credentials. Requests send source content to the endpoints you configure; use only data you are allowed to share.