what's Dolphin?

Dolphin is an AI lab focused on creating uncensored, steerable large language models optimized for distributed inference. Their models, including Dolphin Mistral 24B Venice Edition and variants like Qwen-3.5 integrations, achieve over 4 million monthly downloads on Hugging Face. These models support extended context windows, quantization techniques such as GGUF and FP8 for efficient local deployment, and features like multimodal capabilities via Telegram bots and web UIs. The lab emphasizes synthetic data generation to enable compliant responses to restricted prompts, promoting privacy and user control in AI. Dolphin Network powers decentralized AI inference, allowing users to contribute idle GPUs as nodes to process models and earn POD tokens. The software supports remote monitoring, community betas for testing, and mechanisms to prevent sybil attacks in node participation. Built on Base, it enables verifiable, community-driven compute without centralized intermediaries, aligning with open-source AI principles.

Dolphin serves top uncensored models to Venice, a private inference platform.