What uncensored and abliterated mean
Uncensored is a model classification, not a guarantee that every prompt produces a particular response. Abliteration generally refers to modifying a model to reduce learned refusal behavior. Results still depend on the model, fine-tuning, system prompt and application. Treat the two labels as useful starting points for evaluation rather than interchangeable performance scores.
Kearnix lists these attributes in the model catalog. Kearnix Cloud identifies the service offering for uncensored and abliterated entries; a model's developer is listed separately. Check the model detail page for its available metadata and supported endpoints.
Choose for your actual workload
Start with a small evaluation set drawn from your own application: writing quality, instruction following, coding tasks or conversational consistency. Compare a smaller model with a larger option on the same prompts. Track usable results, latency and tokens consumed, rather than assuming the largest context window or highest price is best.
Use the live catalog to filter model attributes and compare input and output prices. An uncensored label does not establish support for tools, images, structured output or a particular context length. Verify those capabilities on the exact model entry before integrating it.
Make a first request
Create a Kearnix account, add prepaid credits and create an API key. Choose a model ID from the current catalog and make a short test request through /v1/chat/completions. The OpenAI-compatible guide below shows how to configure the base URL and send the request.
Keep the key on your server. Set a sensible output limit and inspect the response before adding longer prompts or automated workflows. Check account usage after your test, and build error handling before sending production traffic.
Evaluate limits before committing
Compare models on your own data and confirm that your intended use follows the applicable service terms, model license and law. Do not infer data retention, hardware ownership, uptime or benchmark performance from an uncensored label. Review the privacy policy and model metadata for those separate questions.
Prices and availability can change. This guide deliberately links to the live pricing catalog instead of quoting a fixed cheapest-model claim that can become outdated.