DeepSeek releases updated R1 reasoning model

Earlier this year, DeepSeek, a Chinese AI startup, released the R1 reasoning model. It took the AI industry by storm, challenging models from OpenAI and quickly drawing global attention. That’s not all; Microsoft also surprised everyone by bringing DeepSeek R1 to Azure and GitHub later on.

Now, the DeepSeek’s updated R1 model has quietly launched on Hugging Face under the MIT license, allowing commercial use.

This new release is a “minor” upgrade, according to a WeChat post from the company. The model repository includes configuration files and parameters, but no formal documentation or description. Still, it’s massive, clocking in at 685 billion parameters, making it far too large for most consumer hardware.

Despite the low-key update, DeepSeek’s R1 continues to spark debate. U.S. regulators have flagged the startup as a potential national security concern, citing the scale and capability of its models. The fact that the DeepSeek’s updated R1 is now openly available to developers will likely intensify that scrutiny.

For now, DeepSeek’s ambitions seem undeterred. The company continues to push boundaries in reasoning-focused AI, with each release hinting at bigger things to come. Not to forget, Microsoft recently released its Phi-4 reasoning model, which it claims is in line with DeepSeek’s R1.