DeepSeek-V4: Efficient Open MoE Models Challenge Closed Labs
Key Questions
What are the key specs of DeepSeek-V4 models?
DeepSeek-V4 includes 1.6T Pro and 284B Flash models with 1M context length and 27% FLOPs savings. They are efficient open MoE models challenging closed labs.
How does DeepSeek-V4 perform on benchmarks?
DeepSeek-V4 crushes coding, reasoning, and agentic benchmarks, rivaling GPT-5.x models. It is adapted for Huawei chips.
What is the context of DeepSeek-V4 amid US-China tensions?
Developed amid $20B+ investments from Alibaba and Tencent, DeepSeek-V4 highlights Chinese AI advances. It is noted in health evaluations.
1.6T Pro/284B Flash 1M context, 27% FLOPs savings; crushes coding/reasoning/agentic benches rivaling GPT-5.x; $20B+ Alibaba/Tencent amid US tensions; noted in health evals.
Sources (2)
Updated Apr 25, 2026