Acree Launches AFM-4.5B AI Model for Enterprises, Ensuring IP Compliance

Acree.ai, a startup dedicated to developing small AI models tailored for commercial and enterprise applications, has announced the release of its latest AI model, AFM-4.5B. This enterprise-focused model is designed to meet the specific needs of businesses while ensuring compliance with intellectual property (IP) regulations. As of July 29, 2025, the model is available for limited free usage by small companies, defined as those with annual revenues under $1.75 million, under a unique 'Acree Model License' that allows them to utilize the model without charge.

The AFM-4.5B model, which boasts 4.5 billion parameters, is significantly smaller than many leading models that can feature tens of billions to trillions of parameters. Its design prioritizes cost efficiency, regulatory compliance, and strong performance, making it suitable for a variety of enterprise applications. According to Lucas Atkins, Chief Technology Officer at Acree.ai, the model is already instruction-tuned for functionalities such as chat, retrieval, and creative writing, ensuring immediate deployment capabilities for enterprises.

Acree.ai's commitment to developing AFM-4.5B stemmed from extensive feedback gathered over the past year from over 150 organizations, ranging from startups to Fortune 100 companies. Many of these organizations reported dissatisfaction with mainstream large language models (LLMs) like those from OpenAI and Anthropic due to their high costs and limited customization options. In contrast, smaller open-weight models, such as Llama and Mistral, while offering more flexibility, raised concerns regarding licensing and geopolitical risks.

To address these issues, Acree.ai has emphasized creating a 'no-trade-offs' alternative that is customizable, compliant, and cost-efficient without compromising model quality. The company's focus on enterprise customer needs is reflected in the rigorous training and development process for AFM-4.5B, which involved curating data meticulously to exclude copyrighted materials and ensure compliance with licensing standards. The model was trained on nearly 7 trillion tokens of data, emphasizing cleanliness and safety in its training process.

The architecture of AFM-4.5B employs a decoder-only transformer design, optimized for performance and deployment flexibility. It features grouped query attention for faster inference and ReLU² activations, which support sparsification without compromising accuracy. Training followed a three-phase approach with pre-training on a wide array of general data, midtraining focusing on math and code, and instruction tuning using high-quality datasets complemented by reinforcement learning methodologies.

Despite its smaller size, AFM-4.5B has demonstrated competitive performance across various benchmarks, averaging a score of 50.13 in evaluation suites such as MMLU and MixEval. It has also shown impressive multilingual capabilities, performing well in over 10 languages including Arabic, Mandarin, and German. According to Acree.ai, the model's modular architecture allows for straightforward support for additional dialects.

The introduction of built-in support for function calling and agentic reasoning further enhances the model's usability, simplifying the process of building AI agents and workflow automation tools, thus reducing the need for complex prompt engineering. This aligns with Acree.ai’s broader strategy to enable enterprises to rapidly develop custom models with lower costs of ownership and easier integration into existing systems.

Looking forward, AFM-4.5B represents a significant advancement in the field of enterprise-ready language models, aiming to provide a solution that balances performance, compliance, and customization. Whether Acree.ai can establish a sustainable position in the rapidly evolving landscape of generative AI will depend on its ability to continue delivering on these promises. Nonetheless, the launch of AFM-4.5B marks a bold step forward in addressing the specific pain points faced by businesses in adopting generative AI technologies.