This article is part of the special issue of the venture bat, “AI’s actual price: performance, performance and ROI scale.” Read more from this special issue.
For the past two decades, businesses have been selected between open source and closed proprietary technologies.
The original choice for enterprises was primarily focused on the operating system, which offered the open source replacement of Linux Microsoft Windows. In the developer’s circle, open source languages ​​such as Azgar and Java script are dominated, as open source technologies, including grants, are standards in the cloud.
The same type of choice between open and closed is now facing businesses for AI, there are numerous options for both models. The proprietary closed model is the largest, most used models on the planet, including Open AI and Entropic. There are models like Meta Lama, IBM Granite, Alibaba’s Kevin and Dippic towards Open Source.
Understanding when to use an open or closed model is an important choice for enterprise AI decision makers beyond 2025 and beyond. There are implications of both financial and customization for either of the two powers that need to understand and consider businesses.
Understand the difference between open and closed licenses
There is no shortage of hyperbal around the old enmity between open and closed licenses. But what does it actually mean for all the enterprise users?
For example, a closed source proprietary technology like Openi GPT4O does not weigh the model code, training data, or model or to see anyone. Available. The model is not easily available for fine and generally, it is only available for the use of real enterprise with a cost (sure, Chat GPT has a free level, But this real enterprise work load lt will not reduce it,
An open technology, like Meta Lalama, IBM granite, or depressic, is openly available. Businesses can usually use models freely, without any restriction, including fine toning and customization.
Rohan Gupta, with a principal DewatetTold Venture Bat that the open vs Source debate is not unique or native to AI, nor is it likely to be resolved at any time.
Gupta explained that closed sources providers usually offer numerous rapper around their model that develops ease, easy scaling, more smooth upgrades and a stable series of down grades and additions. They also provide significant support to the developer. This includes documents as well as hand -related advice and often provide severe integration with both infrastructure and applications. In return, an enterprise plays a premium for these services.
“On the other hand, open source models can provide maximum control, flexibility and customization options, and a dynamic, passionate manufacturer is supported by the environmental system,” Gupta said. “These models are rapidly accessible through a fully organized API in cloud vendors, and expand their distribution.”
To choose between open and closed models for Enterprise AI
The question that many enterprise users can ask is better: an open or closed model? The answer is not necessarily one or the other.
“We don’t see it as a binary choice,” David Gorira, Generate AI Leader EY AmericaTold the venture bat. “Open VS closed is rapidly a fluid design space, where models are selected, or automatically orchestats, based on trading between accuracy, delays, cost, interpretation and security at various locations in the workflower.”
Gorera noted that closed models limit how deep organizations can improve or adopt behavior. The proprietary model shopkeepers often ban fine toning, charge premium rates, or hide this process in black boxes. Although API -based tools simplifies integration, they mostly summarize control, making it difficult to build a very specific or explanatory system.
On the contrary, open source models allow for specific use of cases of targeted targeted targeted targeted, Guardial design and correction. This agent is more important in the future, where models are no longer a common purpose tools, but are the ingredients that exchanged dynamic workflows. The ability to finely shape the model’s behavior at a low cost and with full transparency, when task becomes an important competitive advantage when specific agents or regularly deploying regular solutions.
“In practice, we predict the future of an agent where the selection of the model is summarized,” Garira said.
For example, the user can draft an email with an AI tool, summarize legal documents with the other, find the enterprise documents with an excellent tonic open source model, and locally interact with a device with AI, without knowing what model is known.
“The real question is: which mix of models is in line with the specific demands of your workflow?” Gorira said.
To consider the total cost of property
With the open model, the basic idea is that the model is available for use independently. On the contrary, businesses always pay closed models.
The truth is more important when it comes to considering the total cost of ownership (TCO).
In Parveen Akirjo, Managing Director Visual Partners Venture Bat explained that TCO has many different layers. Some important reservations include infrastructure hosting costs and engineering: Is open source models self -hosted by an enterprise or cloud provider? How much engineering is needed to run the model safely, including Fine Fine Toning, Guard Railing and Security Testing?
Akirajo noted that Fixing an open weight model can sometimes be a very complicated task. Closed Frontier Model Companies make a lot of engineering efforts to ensure performance in multiple tasks. In his view, unless the enterprises deploy similar engineering skills, do not, they will face a complex balance process when fixing the open source model. When organizations choose their model deployment strategy, it produces cost implications. For example, enterprises can fix multiple model versions of different works or use an API for multiple tasks.
Ryan Grass, Head of Data and Applications in Cloud Ancestor Services, Klanint Venture Bat told that from his point of view, the terms of licensing do not matter, except that in the scenario of the Edge case. When the data requirements are available, the biggest restrictions are often related to the availability of the model. In this case, deploying an open model on infrastructure such as Amazon Sage Makers can be the only way to get a sophisticated model that still complies. When it comes to the TCO, Grass noted that the trade of the token is between costs and hosting and rehabilitation costs.
“There is also a clear break, where economics change open models because of cheap,” Grass said.
In his view, for most organizations, closed models, which are resolved by the hosted and scaling organization, will have less TCO. However, for big businesses, sauces companies demand a lot of demand on their LLMs, but simple use issues that require Frontier Performance, or AI Sentric Products companies, can be more costly to hosting open models.
How did an Enterprise Software Developer evaluate Open vs
Josh Boskies, CTOOT The second front system There is a lot of firms that have to consider and assess open VS closed models.
“We use open and closed AI models, depending on specific issues, safety requirements and strategic purposes, we use,” Boskiz told Venture Bet.
Boscoz explained that open models allow his firm to connect modern skills from the beginning without the time or cost of training models. Open, open model of internal experiences or high -speed prototypes, helps to rapidly repeat his firm and take advantage of community -driven developments.
“On the other hand, closed models are our choice when data sovereignty, enterprise grade support and security guarantees are necessary, especially for customer -facing applications or sensitive or regular environment.” “These models often come from reliable shopkeepers, who offer strong performance, compliance support and self -hosting options.”
Boskos said the model selection process is proportional and dangerous, which not only technical fit but also reviews data handling policies, integration requirements and long -term scaleblasticity.
Looking at the TCO, he said that it is significantly different between open and closed models, nor any point of view is globally cheap.
“It depends on the scope of deployment and organizational maturity,” Boskiz said.
What does this mean for Enterprise AI strategy
Smart Tech decision makers are not about to select the open -minded debate, to review the AI ​​investment in 2025. This is about creating a strategic portfolio approach that improves different issues within your organization.
Instant action items are straightforward. First, audit the burden of your existing AI work and map them against the decisive framework described by the experts, which includes accuracy requirements, delays requirements, cost obstacles, safety demands and compliance responsibilities in every use. Second, the model honestly evaluate the engineering capabilities of your organization’s your organization’s engineering, because it directly affects the actual cost of your ownership.
Third, start experimenting with the model orchestration platform that can automatically lead the tasks to the most appropriate model, whether open or closed. It positions your organization for the future of the agent that industry leaders, such as EY Gorira, predict, where the selection of the model is hidden for the closing users.