At Arcee AI, we're committed to democratizing access to cutting-edge AI through open-source collaboration.
Our goal is to empower developers and organizations to build, adapt, and improve state-of-the-art AI models that deliver immediate ROI.
Our open-source models provide a robust foundation for enterprise customers to build and tailor AI solutions specific to their business needs.
A 14B parameter model designed for general-purpose use cases
An 8-billion parameter model distilled from our flagship SuperNova model
A 7 billion parameter model designed for general -purpose use cases
A 7 billion parameter model expertly designed for both Arabic and English
A 7 billion parameter model expertly designed for function-calling and tool-use
A compact 1.5 billion parameter language model for resource-constrained environments
A 72 billion parameter language model for general-purpose use cases
A versatile chat model that specializes in creative writing and functions as a writing assistant
A 70 billion parameter domain-specific model for SEC data analysis
10B-parameter model and excels in a variety of tasks, demonstrating advanced reasoning, code generation, and mathematical problem-solving capabilities.
32B distillation of DeepSeek-V3, which has delivered our highest scores yet on public benchmarks and performs well in advanced chatbots and enterprise data analysis.
You can deploy and test our models on these platforms
Our open-source libraries provide easy-to-use tools for researching, exploring, and enhancing the adoption of advanced model training techniques.
A framework for model distillation, allowing smaller models to learn from larger, more knowledgeable ones.
A library for evolutionary optimization, facilitating the discovery of better models through iterative refinement.
A pre-training methodology for models, enhancing their performance and adaptability.
Check out the latest publications by the Arcee AI team
Learn more about MergeKit, an open-source library enabling efficient model merging to create multitask models without additional training, preserving original capabilities while enhancing AI performance and versatility.
This is a preprint technical report with thorough evaluations to understand the entire process of domain adaptation, continual pre-training, and model merging for enhancing language model performance on financial regulatory data.
This paper examines model merging techniques, from simple averaging methods like Model Soups to advanced approaches like DARE and TIES-Merging, and introduces Differentiable Adaptive Merging (DAM) for efficient model integration.
Subscribe to our newsletter
.svg)
