DeepSeek
AI · Infrastructure
DeepSeek Unveils V4 Preview: 1M Token Context MoE Models
April 24, 2026
DeepSeek launched preview versions of DeepSeek-V4-Pro and V4-Flash, both Mixture-of-Experts models with 1M token context windows, open-source weights, and API access. They lead open models in agentic coding, reasoning, and efficiency on Huawei chips.
Read the Full Story in Velocity
Get curated startup and tech news with new stories published daily, company insights, and real-time alerts.