On Tuesday (20), DeepSeek announced the open-sourcing of DeepInfer, software it uses to run its artificial intelligence (AI) models. The tool was developed to optimize this process, seeking to increase processing efficiency and reduce the time needed to generate responses.
With the publication of the code, developers and companies can study how the program works and adapt it to different projects, while observing the technical requirements and terms of use. According to DeepSeek, the initiative seeks to reduce the use of computational resources and energy, as well as make the execution of AI models more efficient. The results, however, depend on the conditions of use and the available infrastructure.
DeepSeek open-sourced DeepInfer, software that seeks to accelerate the execution of its artificial intelligence models. – Credit: Mojahid Mottakin / ShutterstockIn summary:
- DeepSeek open-sourced DeepInfer, software for running AI models;
- The tool seeks to speed up responses and reduce computational and energy consumption;
- Currently, DeepInfer is compatible with the DeepSeek-V4.1-Flash model;
- The open-source code allows adaptations, but does not include all of the company’s infrastructure;
- The initiative seeks to expand the adoption of the technology and strengthen DeepSeek’s position in the market.
For now, DeepInfer is compatible with the DeepSeek-V4.1-Flash model.Although the opening allows other developers to examine and use the technology, this does not mean that the company’s entire execution system has been made available. Some of the infrastructure components remain outside the open-source code.
Expert uses a car analogy to explain DeepSeek’s move
In an appearance on the Olhar Digital News program, physicist Roberto ‘Pena’ Spinelli, a Machine Learning specialist, analyzed the news. The researcher explained the difference between making an artificial intelligence model available and opening the software responsible for running it. To illustrate this distinction, he used a comparison with a car.
Pena explained that sharing the model is not the same as providing the tools needed to run it efficiently. While the model contains the parameters that guide the generation of responses, the inference software brings together mechanisms that make it possible to put it into operation. Opening this code may facilitate the work of developers who wish to implement their own solutions.
By opening the DeepInfer code, DeepSeek goes beyond the AI model and shares mechanisms for running it more efficiently. – Credit: AI-generated image
“Previously, they shared only the engine blueprint, which is artificial intelligence itself. What they are doing now is delivering the operation of the entire car.“All the software and the mechanism by which they run artificial intelligence were made available to the public,” said the specialist.
The comparison helps explain the difference between the two components, but there is one caveat: the release of DeepInfer does not mean that all elements of the infrastructure used by DeepSeek are available. Nevertheless, access to the code enables researchers and programmers to examine the execution techniques adopted by the company and assess how to leverage them in other projects.
DeepSeek opened the DeepInfer code to developers, but did not make all the technology used by the company available. – Credit: AI-generated image
Opening the code aims to broaden the technology’s reach
The initiative may also support DeepSeek’s commercial strategy. By making part of its technology available, the company creates opportunities for other developers to use it, contribute improvements, and develop compatible applications. This adoption may broaden the reach of its technology in a market characterized by competition among companies seeking to offer more efficient and accessible models.
“DeepSeek wants to create an ecosystem so that more people and companies start using its products,” Pena explained. “By showing how to assemble the full structure, it encourages the public to abandon more expensive paid platforms. It is a consistent way to establish itself in an extremely competitive sector.”
The release of DeepInfer reinforces that competition in the AI market depends not only on the models’ ability to produce responses, but also on the efficiency required to run them.By sharing part of its technology, DeepSeek expands the possibilities for using the software and seeks to strengthen its presence in the sector. The impact of the initiative will depend, among other factors, on the performance achieved in different environments and on the adoption by developers and companies.
Flávia Correia is a journalist at Olhar Digital, covering Science and Space.
