DeepSeek Unveils V4 Series: Open-Source Contender Aims to Challenge Proprietary Giants on Cost and Scale
Large Language Model
DeepSeek V4
Mixture-of-Experts
Open-weight model
Artificial Intelligence
AI Benchmarks
7
Aggressive Open-Source Pricing Pressure
Media Hype
7/10
Real Impact
7/10
What is the Viqus Verdict?
We evaluate each news story based on its real impact versus its media hype to offer a clear and objective perspective.
AI Analysis:
The news represents a genuine competitive challenge (Impact 7) due to cost and scale, significantly raising the open-source bar, despite being standard model iteration news (Hype 7).
Article Summary
DeepSeek has released two preview versions of its new large language model, DeepSeek V4, along with an accompanying R1 reasoning model. These models, V4 Flash and V4 Pro, utilize a Mixture-of-Experts (MoE) architecture and boast massive 1-million-token context windows, enabling complex document and code analysis. V4 Pro is touted as the industry's largest open-weight model (1.6 trillion parameters), significantly outpacing competitors like Moonshot AI and MiniMax. While DeepSeek claims performance comparable to, or surpassing, models like GPT-5.4 and Gemini 3.0 Pro, the company itself notes a performance gap of 3-6 months behind current state-of-the-art frontier models on pure knowledge tests. Crucially, DeepSeek positions itself as a major economic disruptor by offering models that are substantially more affordable than any commercial frontier offering.Key Points
- The V4 series uses a massive MoE architecture with 1M context windows, supporting large-scale enterprise tasks.
- V4 Pro is marketed as the largest open-weight model (1.6T parameters), challenging industry leaders on scale.
- DeepSeek leverages dramatic cost advantages, pricing its services below comparable commercial models like GPT-5.4 and Claude Haiku.