Models: Closed vs Open-Weight vs Open-Source
AI terminology often treats closed, open-weight, and open-source models as interchangeable, although they provide different levels of access, control, and transparency. Model openness is not binary: a model includes components such as trained weights, training data, training code and methodology, and licensing terms, and providers may release some components while retaining others. The trained weights are particularly important because they contain the parameters learned during model training that determine how the model processes inputs and generates outputs. Understanding which components are available is essential when evaluating an AI model.
Closed Models
A closed model does not make its trained weights publicly available, with access typically provided through an API or hosted service. OpenAI’s GPT-5 series, Anthropic’s Claude, and Google’s Gemini are examples of these models. This approach provides access to highly capable models without requiring organizations to operate the underlying infrastructure, while model updates and maintenance remain with the provider. In exchange, organizations have limited visibility into the model’s underlying components and less control over deployment, customization, pricing, and model versioning.
Open-Weight Models
An open-weight model makes its trained parameters available for download, allowing organizations to run and fine-tune the model while the training data, code, and complete training methodology may remain proprietary. Examples include Meta’s Llama, Mistral, Alibaba’s Qwen, DeepSeek, Google’s Gemma, and OpenAI’s gpt-oss, released in August 2025 under the Apache 2.0 license. Open-weight models provide greater control over deployment and customization, including the ability to run models within an organization’s own infrastructure and fine-tune them using proprietary information. However, access to the weights does not necessarily provide transparency into how the model was trained, its data provenance, or all of the conditions governing its use. We previously explored the use of fine-tuned LLMs to embed organizational intelligence.
Open-Source Models
An open-source AI model, under a strict definition, provides the components and information required to study, modify, and reproduce the system. The Open-Source Initiative’s Open-Source AI Definition, published in late 2024, extends beyond model weights to include access to training code, information about the training data, and sufficient documentation to reproduce an equivalent system. Under this definition, many models described as “open” are more accurately classified as open-weight. The Allen Institute for AI’s OLMo family is an example of a more comprehensive approach, providing weights alongside its Dolma training corpus, code, checkpoints, and logs. This level of access enables greater transparency and reproducibility, although fully open models may involve trade-offs in model capability and performance.
The three categories represent different trade-offs between capability, control, customization, and transparency. Closed models provide managed access to model capabilities; open-weight models provide greater control over deployment and customization, and open-source models provide broader access to the components required to inspect and reproduce the system. For enterprises, the relevant question is therefore not which model type is universally preferable, but which parts of the AI stack the organization needs to control. Open-weight adoption is also expanding, with these models accounting for roughly a third of tokens served across major model-routing platforms, while the entry of frontier AI companies into the category is further broadening their adoption, as discussed in https://elaxtra.com/insights/ai-as-an-operating-system.
Elaxtra Advisors is an M&A and value-creation advisory firm that assists institutional investors, private equity-owned platforms, and strategic acquirers invest and create value in worldwide technology services companies. Please contact us to explore potential partnerships.