Sunday, 13 July 2025
27.5 C
Singapore
28.5 C
Thailand
19.9 C
Indonesia
28.1 C
Philippines

DeepSeek updates its powerful AI model and makes it available on Hugging Face

DeepSeek shares an upgraded version of its R1 AI model on Hugging Face, now available under MIT licence with 685 billion parameters.

If you’re following developments in artificial intelligence, you may want to know that Chinese startup DeepSeek has just released an updated version of its R1 reasoning AI model. The company shared the news on WeChat Wednesday morning and the model is now live on the developer platform Hugging Face.

What’s new in the updated model?

DeepSeek describes the latest R1 release as a “minor” upgrade, though it still carries some impressive stats. The model has a massive 685 billion parameters, which are also called “weights.” These key parts guide how the AI performs tasks and makes decisions. With so many parameters, this AI model is far larger than most currently available and is unlikely to run on everyday computers without special hardware.

While the update may be considered minor, it is still notable. The updated R1 model is available under the permissive MIT licence. This means you’re free to use it for commercial projects without paying licensing fees or asking for permission. However, DeepSeek has not yet provided a complete description of how this version of R1 works. The Hugging Face listing only includes the necessary files to run the model, including the configuration details and weight files.

DeepSeek continues to attract global attention

DeepSeek gained wide attention earlier this year when it launched the original R1 model. The powerful AI stood out in the field for its reasoning abilities and performance, directly competing with models made by industry leaders like OpenAI. Many developers and researchers found R1’s results impressive, sparking growing interest in Chinese AI startups.

However, DeepSeek’s rapid progress hasn’t gone unnoticed by government officials. Some regulators have raised concerns that the company’s advanced technology might pose a national security risk in the United States. They worry that powerful AI models like R1 could be misused or create unfair advantages in global tech competition.

Even so, DeepSeek has pressed on, releasing new versions and improvements. By making the R1 model publicly available under an open licence, the company aims for broader global adoption, allowing developers worldwide to test and build with the technology.

What you need to know if you want to use R1

If you’re interested in trying out DeepSeek’s upgraded R1 model, you’ll need access to strong computing power. Due to the many parameters, the model is best suited for research labs, tech companies, or cloud-based environments that can handle such size and complexity.

Although the Hugging Face page does not provide detailed guidance, experienced developers can work with the available files. Getting started may require some technical skill, but the possibilities are wide-ranging—from natural language tasks to complex reasoning and decision-making models.

DeepSeek’s move to share the updated R1 openly might help push the boundaries of what AI models can achieve. Whether you’re a developer, researcher, or just interested in the future of AI, this latest release is another step in the fast-moving world of artificial intelligence.

Hot this week

Singapore to get Huawei’s 480kW ultra-fast EV charger by the end of 2025

Huawei brings 480kW ultra-fast EV charger to Singapore by late 2025, slashing charge times and boosting support for commercial vehicles.

Xiaomi Sound Pocket review: Small in size, big on sound

The Xiaomi Sound Pocket is a sleek, compact speaker with IP67 rating, smart tuning, and strong battery life for all-day listening.

Sunday expands personalised insurance reach with Salesforce support

Sunday expands personalised insurance to over 330,000 users with Salesforce, driving growth and efficiency across Southeast Asia.

Persona 5: The Phantom X finally arrives in Southeast Asia

Persona 5: The Phantom X launches in Southeast Asia with a fresh story, fan-favourite characters, and a special event running until July 31.

Microsoft’s Copilot Plus features may come to desktops by late 2025

Intel’s Arrow Lake Refresh may bring Copilot Plus AI tools to desktop PCs by late 2025, with a focus on enhanced NPU support.

OpenAI preparing to launch AI-powered web browser to rival Chrome

OpenAI plans to launch a new AI-powered web browser, aiming to transform the browsing experience using ChatGPT technology.

Singapore to get Huawei’s 480kW ultra-fast EV charger by the end of 2025

Huawei brings 480kW ultra-fast EV charger to Singapore by late 2025, slashing charge times and boosting support for commercial vehicles.

Samsung, Google, and Qualcomm share their vision for where mobile AI is heading

Samsung, Google, and Qualcomm share how mobile AI will become more helpful, personal, and invisible in your everyday life.

Razer unveils DeathAdder V4 Pro with pro-level features and ultra-lightweight design

Razer’s DeathAdder V4 Pro lands with 8000Hz wireless polling, a lighter design, and esports-level precision for serious gamers.

Related Articles

Popular Categories