Phi-3-medium-4k-instruct API

What is Phi-3-medium-4k-instruct?

Use the public identifier di-microsoftphi-3-medium-4k-instruct-a08fd4b664 for Phi-3-medium-4k-instruct in ServerNet. The gateway uses this identifier to select the configured route; the developer is microsoft.

Where to use Phi-3-medium-4k-instruct

• Add Phi-3-medium-4k-instruct to a support assistant that receives the relevant conversation as text.
• Test Phi-3-medium-4k-instruct on summaries and extraction using your own documents, with a small output limit first.
• Compare answer quality, total token usage and cost per completed task before choosing Phi-3-medium-4k-instruct for production.

Build your integration

Send system and user messages, set max_tokens and store X-Request-Id with your usage record. Stream only when this route supports it. Function calls describe actions for your application to run; the gateway does not execute your tools.

Inputs and service limits

The current chat route accepts text only and one completion per request. Images, audio and hosted paid tools are not enabled by the source model’s capabilities. The configured context and output limits may be below the developer’s maximum. Test language quality on your own dataset.

Connect to Phi-3-medium-4k-instruct API

Enable ai:chat on your key and ai:models:read to check availability. Set SERVERNET_API_KEY in your server environment. Change the idempotency key for each new request; keep it unchanged only for an identical retry.

This example documents the request format. Send it only after this model appears in GET /v1/models.

curl https://servernet.cloud/v1/chat/completions \
  -H "Authorization: Bearer $SERVERNET_API_KEY" \
  -H 'Content-Type: application/json' \
  -H 'Idempotency-Key: CHANGE_FOR_EACH_NEW_REQUEST' \
  -d '{
    "model": "di-microsoftphi-3-medium-4k-instruct-a08fd4b664",
    "messages": [
        {
            "role": "user",
            "content": "Hello"
        }
    ],
    "max_tokens": 128
}'

Use the same Idempotency-Key when retrying an identical request. Successful responses are replayed without a second charge for 24 hours. Do not automatically retry when x-should-retry:false is present. The key is shown once. Store it in a server environment variable. Keep it out of browser code, GitHub and logs.

Compare models for the same task