Venice: Uncensored

Venice: Uncensored is a text model published under the model id cognitivecomputations/dolphin-mistral-24b-venice-edition. It is a fine-tuned variant of Mistral-Small-24B-Instruct-2501, developed by dphn.ai in collaboration with Venice.ai. The provider listed for it is Venice, and the provider release date is 2025-07-09. It sits in the text category and is described as an uncensored instruct-tuned large language model.

The model takes text as input and returns text as output, so it is a straightforward chat and completion model with no image or audio handling. It has a context length of 128,000 tokens, which allows long documents, extended conversations and large prompt payloads to be processed in a single request. The maximum output is 8,192 tokens, so responses are capped well below the input context size. Its knowledge cutoff is 2024-04-30, meaning it has no awareness of events after that date unless they are supplied in the prompt.

Because it is an instruct-tuned model with a large context window, it fits tasks such as long-form question answering over lengthy source material, summarising or extracting information from multi-document prompts, and conversational agents that need to retain a long history. Its uncensored tuning makes it relevant where a developer wants fewer refusals on sensitive or controversial subject matter, for example in fiction writing, roleplay, debate simulation or research into model behaviour. The 8,192 token output ceiling means it is better suited to answers and drafts than to generating very long single-pass documents.

You call it through api14's OpenAI-compatible API using the model id "cognitivecomputations/dolphin-mistral-24b-venice-edition". The supported request parameters are frequency_penalty, max_tokens, presence_penalty, response_format, stop, temperature, top_k, top_p. There is no separate parameter list beyond these, so sampling behaviour is controlled through temperature, top_p and top_k, repetition through the two penalties, and output length through max_tokens. Structured output can be requested via response_format, and generation can be halted on given strings with stop.

Two constraints are worth weighing before you commit. The knowledge cutoff of 2024-04-30 means time-sensitive questions need retrieval or fresh context injected into the prompt. The 8,192 token maximum output limits single-response length, so if you need much longer outputs you would need to chain calls or choose a model with a higher output ceiling. Input is text only, so any task involving images or audio requires a different model.