One of the biggest headlines this week is DeepSeek’s unveiling of its most powerful model variant to date, which is being compared with Google’s new Gemini 3 Pro. The Chinese start-up released two variants on Monday: the base model V3.2, which DeepSeek says is on par with OpenAI’s GPT-5, and V3.2-Speciale, which achieved gold-medal performance on the International Mathematical Olympiad test – a feat previously reached only by internal, unreleased models from OpenAI and Google. The accomplishment has sparked excited discussion, especially since it coincided with the prestigious NeurIPS conference. The most impressive part? DeepSeek achieved this despite having "fewer total training FLOPs" than its US rivals. This is critical because US export controls limit Chinese access to the advanced chips needed for maximum computing power – which is measured in FLOPs, or floating point operations per second. DeepSeek’s efforts underscore how a rising cohort of nimble Chinese start-ups are quickly emerging as strong contenders to US labs. #🇨🇳 ChinA.I. 🤖🧠🦾🤖 https://www.scmp.com/tech..