In a surprising move that has piqued the interest of the AI community, DeepSeek has rolled out a minor update to its R1 model, defying expectations of a larger-scale R2 release.
Announced late yesterday via official channels, the DeepSeek team confirmed a trial upgrade to the R1 model, inviting users to test its capabilities through their official website, apps, and mini-programs. The API interface and usage protocols remain unchanged, streamlining the transition for existing users.
Early testing of the open-source R1 (0528) version has yielded impressive results, particularly in the domain of programming. **Initial evaluations suggest coding proficiency that is now competitive with leading international standards.**
Beyond its prowess in code generation, the updated R1 demonstrates remarkable skill in a variety of tasks. From website generation and basic game creation to the complex task of debugging drawing board applications and implementing image gallery lightboxes, the model exhibits a strong ability to generate, understand, and rectify code. Notably, in certain applications, its performance has even surpassed that of Claude 4.
The R1 update showcases its strength in text processing as well. Demonstrating adeptness at tasks that require specific stylistic knowledge, such as translating culinary terminology or creating stand-up comedy routines, the updated model performs well.
Original article, Author: Tobias. If you wish to reprint this article, please indicate the source:https://aicnbc.com/1183.html
