DeepSeek V4 AI Model Released with 1.6 Trillion Parameters, Challenges US Tech Giants Despite Performance Gap

Chinese artificial intelligence startup DeepSeek released its highly anticipated V4 large language model on Friday, marking its most significant update since a previous release that shook global tech markets over a year ago. The new model features 1.6 trillion parameters and represents the company's first new-architecture model since January 2025. DeepSeek designed V4 as an open-source system, meaning listeners can download, use, and modify it freely. The company claims the model rivals leading closed-source American systems from Anthropic, OpenAI, and Google on major benchmarks while outperforming other open-source competitors. However, according to analysis from the Council on Foreign Relations, V4 does not match the performance of frontier U.S. models. The technical documentation shows V4's reasoning capabilities trail state-of-the-art frontier models by approximately three to six months, with the model explicitly acknowledging it falls short of OpenAI's GPT-5.4 and Google's Gemini 3.1-Pro. Despite this gap, V4 contains significant engineering achievements. The model features a hybrid attention architecture enabling a one-million-token context window at a fraction of the previous V3 model's inference compute cost. This demonstrates Chinese AI laboratories are keeping pace with algorithmic efficiency advancements occurring in closed-source American labs. Within China itself, the competitive landscape has tightened considerably. Moonshot's Kimi K2.6 and Zhipu's GLM 5.1 perform comparably on most public benchmarks, with DeepSeek's V4 Pro pricing higher than either competitor. The release comes amid broader developments in the AI sector. Meta reported its fastest revenue growth since 2021, though the stock dropped three percent following the announcement as rising AI development costs pressured profits. Separately, venture capital firm 137 Ventures raised 700 million dollars to support growth-stage startups focusing on defense and AI sectors, demonstrating continued investor confidence in emerging technology companies. However, questions surround DeepSeek's development methods. The company likely trained V4 using smuggled Nvidia Blackwell chips still banned from export to China under U.S. Commerce Department rules. Additionally, Anthropic and OpenAI have alleged DeepSeek engaged in industrial-scale distillation attacks, creating over twenty-four thousand fake accounts and conducting more than sixteen million interactions to extract capabilities from their Claude models. Thank you for tuning in to this week's AI news briefing. Be sure to come back next week for more updates on the rapidly evolving artificial intelligence landscape. This has been a Quiet Please production. For more information, visit Quiet Please dot A I. Some great Deals https://amzn.to/49SJ3Qs For more check out http://www.quietplease.ai This content was created in partnership and with the help of Artificial Intelligence AI.