Call an uncensored DeepSeek model with curl through an OpenAI-compatible API

#ai #tutorial #api #beginners

Disclosure: Solo Soft builds Solo Router. AI assisted the wording of this tutorial. The offer details below were checked against the live offer page; the sample is not a performance benchmark.

This walkthrough shows the three pieces needed for a first chat-completions request: an API key, a base URL and a model ID. It uses Solo Router's limited free DeepSeek V4.1 Flash Uncensored launch offer, so you can try the request without a payment card while the offer is available.

Offer deadline: November 1, 2026 at 06:58 UTC, or exhaustion of the shared free budget, whichever comes first. This is a fixed launch window, not 48 hours from your signup. Other models are paid.

1. Create a key

Sign in with Google at Solo Router and create an API key. Set it in your own terminal environment as SOLO_ROUTER_API_KEY. The examples below reference that variable rather than embedding a real key in the request.

Base URL: https://solorouter.ai/api/v1
Model: unrestricted/deepseek-v4.1-flash

2. Send a short request

curl https://solorouter.ai/api/v1/chat/completions \
  -H "Authorization: Bearer $SOLO_ROUTER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "unrestricted/deepseek-v4.1-flash",
    "messages": [
      {
        "role": "user",
        "content": "Write a short fictional villain monologue about losing a chess match. Keep it non-graphic."
      }
    ],
    "max_tokens": 256
  }'

The Authorization header carries your API key. The model field chooses which model receives the request. The messages array holds the conversation, and max_tokens asks for a small output so the first test stays focused.

We have intentionally not pasted a fabricated response here. Run the request and inspect the actual result in your own environment.

3. Understand the limits before making a loop

For the free model, each account can use up to 500K tokens a day, resetting at 00:00 UTC. Each request allows up to 32K prompt tokens and 8K output tokens, with a limit of 20 requests per minute. There is also a shared free budget for the offer as a whole.

Keep an initial test sequential. If you receive an error, read the returned status and message before retrying: a bad key, unavailable model, request limit and ended promotion need different fixes. Do not assume that every error means the API is temporarily down.

4. Move to your own client

If your client supports a custom OpenAI-compatible endpoint, configure the same base URL, key and model ID there. Check its documentation for the exact setting names. Matching the API format does not guarantee that every provider-specific feature or every coding client will work unchanged.

Solo Router's broader purpose is to let developers choose supported models from DeepSeek, Kimi, GLM, Qwen, MiniMax, Llama and others using one key and prepaid balance. Outside this limited free model offer, credits start at $5 with usage-based pricing and no subscription. Changing a model ID can change both behavior and price; inspect the model's details before running a large job.

5. Evaluate the result for your task

Use a prompt from a real task, keep an output limit, and check instruction-following, accuracy, repetition and latency. Try more than one request before choosing a default model. This tutorial does not claim that the model outperforms another API.

Uncensored identifies the model variant; it is not a guarantee of zero refusals or permission for any use. Solo Router's AUP prohibits illegal and sexually explicit content.

Offer status, web chat and full limits:
https://solorouter.ai/free-uncensored-deepseek?utm_source=devto&utm_medium=organic&utm_campaign=sr_launch_48h

If you try the custom endpoint in a client, which one did you use and what part of the setup was unclear?

DEV Community