DeepSeek's V4 Models: China's AI Showdown

0:00It's April 24th, Canada News Today starts now. AI-powered and ready. I'm Corey with The Story. Chinese AI startup DeepSeek just launched preview versions of their hotly anticipated V4 models, cranking up the heat in the U.S.-China AI showdown. These open-source pro and flash editions boast major leaps in knowledge, reasoning, and agentic skills, meaning they handle complex tasks, solo, like, bosses. News have been waiting to pit them against heavyweights, such as OpenAI's Chat GPT, Anthropics Cloud, and Google's Gemini. This V4 builds on their V3 from late 2024 and follows the market-shaking R1 reasoning model that dropped in January 2025, proving China can compete on cost and power. Now with a huge 1 million token context window for processing info up from 128,000 in V3, they're more efficient too. The analysts are hyped with one calling V4 super competitive against U.S. rivals, and

1:04another hailing it as a milestone for China's self-reliant tech drive. It's gaining steam and developing nations per Microsoft reports. But U.S. firms, like Anthropic and OpenAI, fired back, accusing DeepSeek of distilling their tech to boost their own, echoing claims from Trump's science, advisor. DeepSeek's own benchmarks show their pro-max version, topping some OpenAI and Google scores, while edging close to the best, like Claw's Opus. The flash holds its own on simpler tasks. Skeptics say it's solid, but no R1 level breakthrough, urging independent checks amid fiercer domestic rivalry. As accusations fly and models evolve, this preview signals China's closing the gapwatch for real-world test to see who leads the pack.