Open-Source DeepSeek V3.2 Just Broke the GPT-5 vs Gemini War!

An open-weight “underdog” model from China is suddenly staring down GPT-5 and Gemini 3—and this time, it brought receipts. DeepSeek V3.2 just arrived with an experimental “Speciale” variant that beats top Western models on multiple reasoning benchmarks, wins four gold medals in elite math competitions, and does it all at API prices so low they’re basically “practically free.” While GPT-5 Pro and Gemini Pro 3 still charge premium rates per million tokens, DeepSeek is offering advanced reasoning, stronger tool use, and near-par agentic behavior for a fraction of the cost—plus the ability to run the weights yourself. In this video, we break down how DeepSeek V3.2 and V3.2 Speciale actually work, why their new “thinking in tool-use” system fixes the classic problem of models losing their train of thought when calling tools, and how an MIT-style open-weight license lets anyone download, modify, and even commercialize the model on their own hardware. We dive into the head-to-head results against GPT-5, Gemini 3 Pro, and other frontier models, explain why governments in the US and Europe are already trying to limit DeepSeek’s reach, and look at how Huawei and other giants are quietly baking it into phones and services. This isn’t just another model release—it’s a direct challenge to the Western AI business model, where ultra-expensive closed systems suddenly have to compete with something fast, cheap, and almost fully open. We’ll show what this means for startups, solo devs, and entire countries that want cutting-edge AI without paying Silicon Valley prices, and whether DeepSeek really is becoming the “alpha and omega” of 2025’s AI arms race. And if you want the real story behind the world’s fastest-moving AI breakthroughs, make sure to like and subscribe to Evolving AI for daily coverage.