🤖Current Models

This section provides a brief description of the models used in our system. These models are categorized into two main types: Text Generation and Image Generation. Note this list may change as we experiment, with new models added and current models potentially being removed.

Text Generation Models

Open-Source Models

Mistral-8x7b Instruct

Mistral-8x7b Instruct is an open-source language model known for its ability to follow instructions and generate contextually relevant text. This model is particularly effective in scenarios where specific guidance or direction is provided, allowing for tailored text generation.


Mistral-7b is a fully open-source Large Language Model from Mistral lab with an excellent potential to fine-tune over custom datasets.


Llama-2-70b is an open-source large language model originally created by Meta. This variant has 70 billion parameters. It excels in understanding and generating text across a wide range of topics and styles. The model is notable for its scalability and adaptability to various text generation tasks.


A smaller version of the Llama model, this 12 billion parameter model can be faster than its larger sibling at the expense of quality that becomes noticeable in replies to complex tasks and queries.

Fine-Tuned Variants

DecentAI leverages some fine-tuned variants of the models above for specific purposes. For example, a fine-tuned version of Mistral is one of the experimental LLMs used for certain kinds of routing between models.

OpenAI Models


GPT-4-1106-Preview represents an advanced stage in the GPT series, offering improved capabilities over its predecessors. As a large-scale language model, GPT-4 is designed to understand and generate human-like text, exhibiting a deeper understanding of context and nuances. The '1106-preview' version indicates a specific iteration of the model that has been previewed for certain functionalities.


GPT-3.5-Turbo-1106 is a variant of the GPT-3.5 model optimized for faster response times and efficient text generation. Developed by OpenAI, it is part of the Generative Pre-trained Transformer (GPT) series and is fine-tuned to provide high-quality text outputs with lower latency. This model is suitable for applications requiring quick and reliable text generation.

Anthropic Models


Claude-2.1 is another iteration from Anthropic, reflecting enhancements in natural language understanding and generation. It is known for its sophisticated conversational abilities and is engineered to handle complex dialogues and queries with a higher level of accuracy compared to earlier versions.


Claude-Instant-1.2, created by Anthropic, is a conversational AI model designed for real-time interactions. It emphasizes speed and responsiveness, making it suitable for applications that require immediate feedback. The model balances the need for quick output with maintaining a high quality of text generation.

Image Generation Models

Dall-E-3 and Dall-E-2

Dall-E 3, developed by OpenAI, represents the latest iteration in the Dall-E series, which is specialized in image generation. This model is capable of creating highly detailed and creative images based on textual descriptions. It leverages advanced algorithms to understand and interpret visual concepts from text inputs, resulting in unique and high-quality image outputs. Dall-E-2, the earlier version, is also useful for certain image styles and resolutions.

Stable Diffusion XL - 2.1

SDXL (Stable Diffusion XL) is an advanced image generation model that builds upon the principles of the Stable Diffusion series. It is designed to produce high-resolution, detailed images and is particularly effective for complex image generation tasks. SDXL combines the power of deep learning with advanced diffusion techniques to create visually striking and accurate images.

Fine-Tuned Variants

Specific image model tunings may lead to better results for specific styles, such as icon or avatar generation.

Last updated