A small evaluation notebook for an uncensored AI API

Published by UncensoredLabs.

A useful model evaluation starts with a notebook of observations, not a winner chosen in advance. Pick three tasks you actually expect your application to handle: a factual question with a checkable answer, a coding task you can run, and a creative task with explicit constraints.

Record the prompt, selected model, output, what you checked and what you would change. Keep response style separate from correctness. An answer that proceeds without refusal can still need correction; a refusal can also be appropriate for your application. This article does not report a controlled benchmark.

When trying UncensoredLabs, use its OpenAI-compatible API with a compatible client, begin with one model and one short request, and inspect usage before broadening the test. The service offers uncensored and abliterated models, streaming chat completions, prepaid token billing, API-key spending limits and model allowlists.

New accounts receive a one-time $0.10 promotional usage credit. Normal model prices and request limits apply; the credit has no cash value and cannot be withdrawn. The amount of testing depends on the model and request usage.

Current models and pricing: https://uncensoredlabs.dev/

Documentation: https://uncensoredlabs.dev/docs

Terms: https://uncensoredlabs.dev/terms

Generated editorial illustration; no benchmark results are depicted.

Earlier public landing layout; see the live site for current prices.