The new update of Deepseek R1 was officially unveiled + video

Chinese startup Deepseek Working as a serious competitor for American artificial intelligence companies, the new discipline model update R1 He introduced himself through the WeChat messenger and then released on the Hugging Face platform.

Deepseek, which attracted a lot of attention this year with the R1 model, has now released a new version of the model with an open MIT license on the Hugging Face platform. This license allows users to use this model commercially.

According to Deepseek, this is a minor update and will be improved compared to the previous version. However, the repository of this model in the Hugging Face lacks accurate descriptions and only includes configuration files and model weights; That is, the same parameters that determine the model’s behavior.

The new R1 model has 685 billion parameters, indicating its huge size. This high volume makes it impossible to run the model without specific changes on home hardware.

Performance of the new Deepseek R1 model of user language

According to data released by some users in the field of artificial intelligence models, the new version of Deepseek R1 has recorded better performance than its previous version in Thematic Generation benchmark. This model’s score in this test has been reduced from 1.80 to 1.74, indicating improvement in the model’s ability to produce subject -oriented content; Because in this criterion, the lower points indicate higher quality.

Another user has also claimed that this version has been significantly improved in the realm of logical reasoning by publishing the following video showing the performance of Deepseek-R1-0528 (the new version) against Claude-4-SONNET. According to the user, the new model is in the field of franz development, and the use of the tools has been more intelligent.

These upgrades can make the R1 a stronger option for developers and professional users.

Deepseek earlier this year made a lot of noise with the introduction of the initial version of the R1 and even compared to big competitors like Openai. However, the technology of this Chinese startup has also attracted the attention of some American regulatory agencies, and some have considered it a threat to US national security.

The move also led to a new wave of models by Chinese companies such as Alibaba and Zhipu AI.