DeepSeek releases V4 AI model with 10x longer text processing — open-source Chinese competitor challenges Western AI dominance

MIT Technology Review AI · April 24, 2026

AI Summary

•Chinese AI company DeepSeek unveiled V4, a new flagship large language model (AI that understands and generates text), on Friday. The model can process much longer prompts than its predecessor — meaning it can read and understand entire documents, codebases, or research papers in a single request instead of splitting them into chunks.
•V4 uses a new architecture that handles large amounts of text more efficiently without requiring proportionally more computing power. This translates to faster responses and lower costs for anyone running the model — a meaningful advantage over competitors that require expensive hardware to process long documents.
•Because DeepSeek releases V4 as open source (free code anyone can download and modify), it becomes a viable alternative to closed, paid models from OpenAI and Anthropic. Software developers and companies now have a free option for building AI features into their products, which could compress margins for expensive commercial AI APIs and force pricing changes across the industry.
•V4 is available now as a preview download for developers; full release details and access links are on DeepSeek's official channels.

Get curated AI news from 200+ sources delivered daily to your inbox. Free to use.