DeepSeek has introduced a major upgrade to its flagship reasoning model, introducing DeepSeek-R1-0528. Now live on Hugging Face, the updated version shows substantial progress in reasoning and inference capabilities, thanks to enhanced computational backing and algorithmic post-training optimizations.

According to benchmark results shared by the company, DeepSeek-R1-0528 now matches the performance levels of OpenAI’s o3 and Google’s Gemini 2.5 Pro across multiple AI evaluations.

In the AIME 2025 test, the new model secured an impressive 87.5%, closely trailing OpenAI’s o3 at 88.9% and surpassing Gemini 2.5 Pro’s 83.0%. In LiveCodeBench—a coding proficiency benchmark—R1-0528 matched Gemini 2.5 Pro with a 77% score, just below OpenAI’s o3 at 78%. On the MMLU-Pro benchmark, a standard for reasoning and general knowledge, the DeepSeek model scored 85%, again aligning with OpenAI’s o3 and Gemini 2.5 Pro.

Users on social media have reported smooth local deployments, including Ivan Fioravanti, CTO at CoreView, who noted running the 4-bit version at 21 tokens per second on an Apple M3 Ultra chip.

Originally launched last year, the DeepSeek R1 model disrupted the AI landscape with its exceptional benchmark scores. DeepSeek’s design philosophy emphasizes efficiency, optimizing model architecture to boost performance without excessive hardware demands.

This aligns with the performance of its earlier model, DeepSeek V3, which achieved unprecedented results using just 2048 NVIDIA H800 GPUs—sparking debates in the tech world after former OpenAI researcher Andrej Karpathy highlighted that such capability typically demands far more hardware. This revelation reportedly contributed to a market cap drop of over $500 billion for NVIDIA.

Also Read: Elon Musk’s xAI secures $300M from Telegram

DeepSeek’s models are widely deployed across sectors, from startups to defense and automotive giants. BMW recently announced plans to embed the model into its vehicles in China. According to reports, it’s also being used by Chinese officials in aviation R&D, legal systems for rapid document drafting, and by healthcare professionals to assist in locating missing individuals.

The momentum behind DeepSeek indicates a strong shift toward open-source, high-performance AI tools that prioritize intelligent architecture over brute-force computing.

Be a part of Elets Collaborative Initiatives. Join Us for Upcoming Events and explore business opportunities. Like us on Facebook , connect with us on LinkedIn and follow us on Twitter.

"Exciting news! Elets technomedia is now on WhatsApp Channels Subscribe today by clicking the link and stay updated with the latest insights!" Click here!