Woter AI detection.Hurry - ends Jul 22nd

New Year Sales :up to 80% OFF

AI Humanize AI Translator Bypass AI AI Rewriter AI Detector

PRICING

TRY FOR FREE

Open source vs. closed source, is DeepSeek the best choice?

Written by

Caleb Hayes

Updated on:July-15th-2025

DeepSeek is driving an open source craze.

On February 14, Baidu announced that the Wenxin model series will be open source on June 30.

Shortly after Baidu open-sourced its code, OpenAI CEO Sam Altman also launched a poll on the X platform, asking netizens what the next open-source project would be. This news revealed that OpenAI also intends to open-source a large model project.

Before this, both Baidu and OpenAI were staunch supporters of closed-source large models. However, driven by DeepSeek, open source has become a trend and the choice of more large-model companies.

But this also brings a problem to enterprises that use big models: what is the difference between open source and closed source? In customer service scenarios, how should we choose between open source and closed source?

The so-called open source big model usually refers to a big model released under an open source protocol, which allows the source code of the software to be freely used, studied, modified and distributed by the public, such as DeepSeek.

Correspondingly, closed-source big models refer to big models whose source code and internal working mechanisms are not open to the public. For example, if companies want to use related capabilities of GPT-4, they need to apply to OpenAI and pay a certain fee.

Regarding open source big models, the biggest perception is probably that they are “free”, which is also one of the most competitive aspects of open source big models.

Let’s take an example of deploying a customer service system in a privatized manner.

When using open source large models, companies need to pay a lot of upfront costs, including purchasing/renting servers, building and maintaining data centers, and hiring relevant personnel to tune and deploy models.

However, these costs only exist in the early stage. After the relevant construction is completed, the enterprise does not need to pay for the subsequent use of the model. However, if a closed-source large model is used, in addition to the initial investment, the model must be paid for according to the token usage in the future.

If the customer service system is deployed on a public cloud or third-party platform.

When using open source large models, enterprises do not need to pay for the model itself, because it follows the open source agreement and the model is free. Enterprises only need to pay cloud vendors for cloud services, such as computing resources (such as CPU, GPU), storage, data transmission, and possible additional services (such as load balancing, automatic expansion, etc.).

When using closed-source large models, companies also need to pay for the models. For example, the input price of the OpenAI-o1 model is US$15/million tokens, and the output price is US$60/million tokens.

In summary, when using open source big models, you do not need to pay for the model, you only need to pay for related costs such as computing power and storage; while closed source big models may add model licensing or usage fees on top of related costs.

That is, in customer service scenarios such as finance, e-commerce, and banking, if the number of customer inquiries is large enough, the long-term use cost will be lower after private deployment using open source large models. However, if the number of customer inquiries is small, it will be more cost-effective to use the public cloud.

For enterprise applications, the key is to find the balance.

So, considering only the model usage effect, are open source big models better than closed source big models? The answer is not necessarily.

From the perspective of application companies, open source big models have the following advantages in addition to cost:

Advantages of open source big models

1. Transparency

The source code of open source large models is public, so companies can understand how the models work and increase their trust in the model's behavior.

2. Free customization and modification

Enterprises can freely modify and customize open source big models as needed to adapt to specific business scenarios and requirements.

3. Avoid vendor lock-in

Enterprises can freely modify and customize open source big models as needed to adapt to specific business scenarios and requirements.

However, open source big models also have some problems, such as stability and reliability issues, lack of technical support and maintenance, and integration, customization, security compliance, and performance optimization all require additional work.

In contrast, the advantages of closed-source big models are:

Advantages of closed-source large models

1. Professional service support

Closed-source large models are usually provided by professional companies, which means that enterprises can obtain more professional customer service and technical support, which is especially important for solving complex problems or customized needs.

2. Performance optimization

Closed-source models may be deeply optimized for specific tasks or hardware platforms, and thus may outperform general-purpose open-source models.

3. Stability and reliability

Closed-source models are typically rigorously tested and validated to ensure stability and reliability in commercial environments.

4. Continuous updates and improvements

Developers of closed-source models may continue to update and improve them, and users can regularly get new features and performance improvements without having to maintain and upgrade them themselves.

5. Security and privacy protection

Closed-source models may provide more advanced security features and privacy protection measures because their design and implementation details are closed to the outside world, reducing potential security risks.

5. Others

Closed-source big models also have advantages in compliance, commercial licensing, integration, and customization

At the same time, closed-source large models also have shortcomings, such as low transparency, high cost, limited customization, and technology lock-in.

Of course, the open source model is not what many people understand, that as long as it is open source, you can use it at will. There are currently seven mainstream open source protocols, some of which are very loose, allowing users to use, modify and distribute software almost without restrictions, such as the MIT License, Apache License 2.0, etc.

However, there are also some open source agreements that have strict restrictions on the commercial use of related codes, such as the GNU General Public License (GPL) agreement, which requires that any work that modifies and distributes GPL software must also be released under the GPL agreement. That is, if an enterprise uses its open source projects, then the enterprise's related applications must also be open source, which will lead to challenges to the enterprise's business secrets and application security.

Therefore, choosing the right open source project under the right open source agreement is also something that companies need to carefully consider when applying the open source big model.

In the customer service scenario, Tianrun Rongtong will take the lead in launching the Weiteng big model platform in 2023. At present, Tianrun Rongtong Weiteng big model platform has access to open source big models such as DeepSeek and Tongyi Qianwen, as well as closed source big models such as Doubao, Kimi, and Wenxin Yiyan.

Tianrun Rongtong Weiteng big model platform relies on innovative big model gateway technology to achieve one-click switching of various underlying big models, greatly improving the flexibility and convenience of model application, and can provide customers with diversified solutions.

As a commercial-grade platform for enterprises, Tianrun Rongtong has been deeply engaged in customer contact for nearly 20 years. It has not only accumulated a lot of know-how in customer contact scenarios in multiple industries, but also has rich experience in model fine-tuning, private deployment, etc., and can escort the implementation of customers' commercial applications.