Uncensored

The Dolphin family, published by Cognitive Computations, is a well-known line of open-weights derivatives that build on Mistral base models and apply community-driven fine-tuning to lower default refusal behavior. The Venice Edition continues that tradition at the 24-billion-parameter scale, combining Mistral-family architecture with a post-training recipe aimed at long-context chat and instruction following rather than conservative safety tuning. Because the open-weight checkpoints are freely redistributable, developers can host the model locally, run inference on their own hardware, or route requests through aggregator providers that expose a hosted endpoint.

In practical terms, the Venice Edition is positioned for users who want a permissive conversational model with a generous context window for extended documents, roleplay, or coding sessions, and who are willing to layer their own policy controls around it. The trade-off is the typical one for "uncensored" derivatives: broader willingness to engage with sensitive or gray-area prompts, but no guaranteed guardrails against harmful completions, so deployment should assume external filtering, human review, or sandboxed environments. It fits best for experimentation, prototyping, and local development workflows where the open-weight format and Mistral lineage matter more than enterprise-grade safety guarantees.