Open-source · Uncensored · Steerable

Models

Dolphin Models

General purpose AI models that give you full control. No arbitrary restrictions. No hidden agendas.


Why Open-Source

The Problem

Why closed models present real issues for businesses and developers.

System Prompt Control

Closed providers maintain control, deprecating and changing things as they wish, often breaking software.

Model Version Control

They change things silently or deprecate models your business relies on without warning.

One-Size-Fits-All Alignment

Alignment is not tailored to your application; it's designed for the lowest common denominator.

Data Privacy Concerns

They can see all your queries and potentially use that data in ways you wouldn't want.


Principles

Our Philosophy

Dolphin belongs to YOU. It is your tool, an extension of your will.

Your Tool, Your Rules

Dolphin is steerable and gives control to you. You set the system prompt. You decide the alignment. You control your data.

No Arbitrary Restrictions

Unlike closed models, Dolphin doesn't impose ethics or guidelines on you. You are the one who decides the boundaries.

Personal Responsibility

Just as you're responsible for what you do with any tool, you are the creator and originator of any content you generate.


Featured

Our Models

Explore Dolphin's collection of uncensored, steerable AI models.

Dolphin X1 Llama 3.1 405B

Dolphin X1 Llama 3.1 405B

406B Parameters Flagship Latest

Dolphin X1 405B directly uncensors Llama-3.1-Tulu-3-405B while maintaining instruction-following capabilities.

View on Hugging Face
Dolphin X1 Trinity Nano

Dolphin X1 Trinity Nano

6B Parameters Arcee Trinity Nano Text Generation

Dolphin X1 Trinity Nano directly decensors Arcee AI's Trinity Nano model while maintaining instruction-following capabilities.

View on Hugging Face
Venice Role Play Uncensored

Venice Role Play Uncensored

24B Parameters Roleplay Instruction-tuned

A 24B generalist roleplay/adventure model built on the Venice Edition base, designed to retain strong instruction-following capabilities.

View on Hugging Face
Dolphin Mistral 24B Venice Edition

Dolphin Mistral 24B Venice Edition

24B Parameters Venice.ai Collab Most Uncensored

A collaborative project between Dolphin and Venice.ai designed to create an uncensored version of Mistral 24B optimized for the Venice ecosystem.

View on Hugging Face
Dolphin X1 8B

Dolphin X1 8B

8B Parameters Llama 3.1 Base Uncensored

An 8B parameter model created by uncensoring Llama 3.1 8B Instruct while maintaining or improving its capabilities.

View on Hugging Face

Dolphin Xgen-RL

RL Optimized Enhanced Reasoning

Reinforcement learning optimized model for enhanced reasoning and instruction following.

View on Hugging Face
Dolphin 2.9.1 Yi-1.5 34B

Dolphin 2.9.1 Yi-1.5 34B

34B Parameters Yi-1.5 Base Function Calling Agentic

An uncensored model trained on instruction, conversational, and coding data with agentic abilities and function calling support.

View on Hugging Face
Dolphin Llama 3.1 8B Instruct

Dolphin Llama 3.1 8B Instruct

8B Parameters Uncensored Fast Inference

Our effort to directly uncensor Llama's 3.1 8B instruct-tuned model. Fast, capable, and free of arbitrary restrictions.

View on Hugging Face

Thank You

Our Sponsors

Crusoe Cloud

16x L40s for training and evals

Akash

On-demand 8x H100 for training

Lazarus

16x H100 for training

Cerebras

Fast inference services for data labeling

Andreessen Horowitz

Grant that made Dolphin 1.0 possible

You

Running nodes, using models, and building the future

Experience Dolphin

Try our models through the web chat or Telegram bot. No restrictions, no censorship.