DeepSeek V4 aims to match frontier AI performance at a fraction of the cost

DeepSeek has released preview versions of its new V4 Flash and V4 Pro models, claiming major efficiency gains and near-parity with frontier systems like GPT-5.4 and Gemini 3.1 Pro on reasoning and coding benchmarks. The flagship V4 Pro uses a massive 1.6 trillion parameter mixture-of-experts architecture, making it the largest open-weight model to date, while both versions offer 1 million token context windows. Despite trailing top US models in knowledge tasks by several months, DeepSeek positions V4 as dramatically more affordable than competing frontier systems. The launch comes amid US accusations that China is stealing AI IP at scale, with DeepSeek previously accused of distilling models from OpenAI and Anthropic.

Read the full story on TechCrunch →