
Open source vs. closed models – what it means for Saga
The legal sector is at the forefront of technological evolution, and artificial intelligence plays a significant role in reshaping how legal professionals work. A critical discussion within this transformation is the choice between open-source and closed AI models. At Saga, understanding the differences between these approaches helps us make strategic decisions in developing solutions tailored to the legal industry.
Open-source models: flexibility and community-driven development
Open-source AI models are characterized by transparency and accessibility. When we talk about "open source" models in AI, it typically doesn't mean access to source code in the traditional software sense. Instead, it usually refers to access to:
- The model weights - these are the numerical parameters that define how the model processes information, often released as large files
- Model architecture - the high-level design and structure of the neural network
- Training methodology - documentation about how the model was trained
- Licensing - permissive licenses that allow others to use, modify, and redistribute the model
For example, Meta's LLaMA or Stability AI's StableLM are considered "open source" because their weights and architectures are publicly available, but you wouldn't see traditional programming source code like you would with open-source software like Linux or Firefox.

For Saga, the potential of open-source models lies in their flexibility. By leveraging open-source frameworks, we can create more transparency, and we can customize models to build functionalities into our platform, ensuring they align with the unique demands of the legal profession.
However, open-source models face several key operational challenges. Hosting these models requires substantial computing infrastructure. Additionally, organizations must implement robust security measures to protect against unauthorized access and potential attacks. Privacy concerns are also significant - organizations need to ensure proper data handling, and maintain compliance with various privacy regulations while running these models. Many organizations find themselves unprepared for the complexity of securely deploying and maintaining open-source models at scale.
Some well-known open-source models are LLaMa (Meta), Mixtral (Mistral), and Granite (IBM).
Closed models: controlled environments and enhanced security
Closed AI models, often developed by private companies, offer a controlled environment with pre-configured capabilities. These systems are typically optimized for specific tasks and are supported by robust infrastructure and security measures.
For Saga, closed models provide a stable foundation for delivering reliable and secure AI functionalities. These models are also backed by consistent support and updates from the providers, ensuring their reliability over time.
Some well-known closed source models are GPT 4o (OpenAI, don’t be confused by the word Open), Claude 3.5 Sonnet (Anthropic) and Gemini (Google).
Saga’s approach: a balanced perspective
At Saga, we recognize that neither open-source nor closed models offer a one-size-fits-all solution. Instead, we take a balanced approach, leveraging the strengths of both to develop a platform that meets the highest standards of security, efficiency, and customization.
At the moment, Saga offers closed source models within the AI platform. However, Saga is about to embrace the hybrid offering soon. This hybrid strategy allows Saga to stay at the cutting edge of legal technology, providing solutions that are not only effective but also aligned with the practical requirements of legal professionals. Every LLM has his own specialty to solve particular cases. By carefully evaluating the trade-offs between open and closed models, we continue to deliver an AI platform that supports our clients in optimizing their operations while maintaining the integrity and confidentiality of their work.
