Cohere Rerank v5 (Pro & Fast) for C# and .NET 10 on Microsoft Foundry

• 📖 Full walkthrough: Cohere Rerank v5 Pro and Fast for C# Developers
• ⭐ If this saved you time, star the repo so other .NET developers can find it.
• For more C# Samples using other AI Models check out AI models for C# developers
• For my Free Ebook on Prompt Engineering check out Prompt Engineering for .NET Developers

A .NET 10 console app that calls Cohere Rerank v5 Pro and Fast on Microsoft Foundry. It uses HttpClient and Entra ID (Azure Identity) authentication, with no SDK. The app ranks a set of passages against one question, compares the two models, and uses a relevance threshold to decide which passages go to a chat model.

Why Rerank v5 Matters

• Reorders vector or keyword search results by real relevance to the question, which improves RAG answers
• Returns a calibrated relevance_score, so you can drop weak passages instead of always passing the top N
• When no passage clears the threshold, the app answers "I do not know" instead of guessing
• Pro gives the most accurate ranking. Fast gives lower latency for high-volume or interactive workloads.

Prerequisites

• .NET 10 SDK
• An Azure subscription with a Microsoft Foundry resource
• Cohere-rerank-v5-pro and Cohere-rerank-v5-fast deployed on that resource
• Either an API key, or Azure CLI signed in with az login (used by DefaultAzureCredential locally)
• For Entra ID sign-in, your identity needs the Cognitive Services User role on the Foundry resource

Configuration

Settings are read from user secrets first, then environment variables.

dotnet user-secrets init
dotnet user-secrets set "COHERE_RERANK_PRO_ENDPOINT" "https://<your-foundry-resource>.services.ai.azure.com"
dotnet user-secrets set "COHERE_RERANK_FAST_ENDPOINT" "https://<your-foundry-resource>.services.ai.azure.com"

# Optional: use an API key instead of Entra ID
dotnet user-secrets set "COHERE_RERANK_PRO_API_KEY" "<your-api-key>"
dotnet user-secrets set "COHERE_RERANK_FAST_API_KEY" "<your-api-key>"

Setting	Required	Purpose
COHERE_RERANK_PRO_ENDPOINT	Yes	Base URL of the Foundry resource that hosts the Pro deployment
COHERE_RERANK_FAST_ENDPOINT	Yes	Base URL of the Foundry resource that hosts the Fast deployment
COHERE_RERANK_PRO_API_KEY	No	API key for Pro. If it isn't set, the app signs in with Entra ID.
COHERE_RERANK_FAST_API_KEY	No	API key for Fast. If it isn't set, the app signs in with Entra ID.

Important: Set the endpoint to the resource root only, such as https://<resource>.services.ai.azure.com. Do not add /models. The app appends /providers/cohere/v2/rerank, which is where Foundry serves Cohere's native Rerank API. With /models in the endpoint, the request returns 404 Resource not found.

Note: Each run calls both models, so set both endpoints. Pro and Fast can be on the same resource or on different ones.

Usage

dotnet run

The app sends one question and six passages to each model. For each model, it prints the response time and the top 3 passages with their scores. Each passage is marked keep or drop based on the model's threshold. Sample output:

Pro (2331 ms, threshold 0.50)
  0.81  keep  Annual plans can be refunded in full within 30 days of purchase. After 30 days, the remaining months are prorated.
  0.13  drop  Refund requests are submitted from the billing page under Invoices.
  0.06  drop  Monthly plans can be cancelled at any time. The change takes effect at the end of the billing period.
  1 passage(s) passed to the chat model.

Fast (2089 ms, threshold 0.45)
  0.81  keep  Annual plans can be refunded in full within 30 days of purchase. After 30 days, the remaining months are prorated.
  0.06  drop  Refund requests are submitted from the billing page under Invoices.
  0.05  drop  Monthly plans can be cancelled at any time. The change takes effect at the end of the billing period.
  1 passage(s) passed to the chat model.

What the sample shows

Concept	Where	Details
Native Cohere v2 request	RerankAsync	POST {endpoint}/providers/cohere/v2/rerank with model, query, top_n and documents
API key or Entra ID	RerankAsync	Uses a bearer API key if one is set. Otherwise it gets a token for https://cognitiveservices.azure.com/.default.
Per-model threshold	ReadModel	Pro uses 0.50 and Fast uses 0.45. Tune both against your own data.
Grounded fallback	Main loop	If no passage clears the threshold, the app answers "I do not know" instead of passing weak context to the chat model
Fixed Content-Length	RerankAsync	Uses StringContent instead of JsonContent, so the request isn't sent with chunked encoding

Troubleshooting

Error	Likely cause
404: Resource not found	The endpoint includes /models or another path. Use the resource root only.
404 / DeploymentNotFound	The model value (Cohere-rerank-v5-pro or Cohere-rerank-v5-fast) doesn't match a deployment name on the resource
401 Unauthorized	The API key is wrong, your identity doesn't have the Cognitive Services User role, or Azure CLI is signed in to the wrong tenant
Name or service not known	The endpoint host is wrong, or DNS or a VPN is blocking it
InvalidOperationException: Set COHERE_RERANK_..._ENDPOINT	A required setting is missing

Note: Environment variables override user secrets. If a value you set is ignored, look for an old environment variable with the same name.

Project structure

Program.cs                              Entry point, configuration, and rerank call
Cohere-Rerank-v5-CSharp-Sample.csproj   .NET 10 project and package references
.github/workflows/dotnet.yml            Build workflow that runs on every push and pull request
CONTRIBUTING.md                         How to report bugs and contribute changes
LICENSE                                 MIT license
README.md                               This file

License

This project is licensed under the MIT License. See LICENSE for details.