DeepSeek-R2 comes with a new chip: Will it be better than GPT-5?

China -based artificial intelligence startup DeepSeek is back on the agenda with its new large language model (LLM), DeepSeek-R2, which it plans to release soon.

Wider Logical Thinking

The model, powered by Huawei's Ascend 910B chipset, is expected to be unveiled between August 15th and 30th. The system, boasting up to 512 PetaFLOPS of processing power, will offer improved logic, responsiveness, and task efficiency compared to the previous R1 model. The company aims to maintain cost-effectiveness, open-source architecture, and efficiency.

NVIDIA SHARES HAD A HISTORICAL LOSS OF VALUE

DeepSeek shocked the US tech world with its R1 model, unveiled in January. DeepSeek-R1, the most downloaded free app on Apple's US app store, rivaled OpenAI's multi-billion dollar GPT models by claiming only $5.6 million in development costs. This led to the largest single-day decline in Nvidia stock in history.

Security concerns in the United States

The R1's success challenged the notion that "bigger is always better," demonstrating that smaller yet efficient models could still deliver high performance. However, the company faced security concerns in the US due to its data centers in China and potential ties to the Beijing government. Experts say that while DeepSeek aims to re-enter global competition with the R2, high-performance chip supply shortages and increasing competition in both the US and China are making it difficult for the company to maintain its momentum.