Dolphin Mixtral
Dolphin Mixtral is a fine-tune of Mistral AI's Mixtral mixture-of-experts base model, created by Eric Hartford as part of the Dolphin series. The defining characteristic of this series is the deliberate removal of refusal and alignment guardrails, producing a model that complies with a broader range of requests without built-in content restrictions.
This makes Dolphin Mixtral a popular choice for researchers and developers who need a capable, uncensored assistant for testing, red-teaming, or creative fiction use cases where standard safety filters would be counterproductive. The underlying Mixtral backbone provides strong language quality, and the Dolphin fine-tune preserves general instruction following ability.
Key Features
Uncensored fine-tune with alignment guardrails deliberately removed
Built on Mixtral's MoE backbone for strong general language quality
Capable multi-turn dialogue and instruction following
Suitable for creative fiction, roleplay, and open-ended storytelling
Useful for red-teaming and safety research scenarios
Ideal Use Cases
AI safety research and red-teaming exercises
Unrestricted creative writing and narrative fiction
Developer testing requiring responses without refusals
Roleplay and character-driven interactive fiction
Research into model behavior without RLHF alignment
Example Prompts for Dolphin Mixtral
Technical Specifications
| Provider | Dolphin |
| Category | Text |
| Modality | Text -> Text |
Frequently Asked Questions
Try Dolphin Mixtral now
Start using Dolphin Mixtral instantly — 100 free credits, no credit card required. Access 750+ AI models through one platform.