DeepSeek has released its V4 series of Mixture-of-Experts models, including V4-Pro (1.6T total parameters) and V4-Flash (284B total). Both models are released under the MIT license, offering full open weights and supporting a context window of up to 1 million tokens. While V4-Pro boasts frontier-class benchmarks, particularly in coding, its large size makes it suitable for datacenter deployment, whereas V4-Flash is more accessible for local use. This release coincides with DeepSeek's significant funding round, reportedly around $7-10 billion, with a stated commitment to continued open-source model releases. AI
IMPACT Sets a new standard for open-weight models with frontier-class performance and a permissive license, potentially accelerating enterprise adoption of self-hosted LLMs.
RANK_REASON Frontier-lab model release with system card and open weights. [lever_c_demoted from frontier_release: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
