DeepSeek previews new AI model that ‘closes the gap’ with frontier models

DeepSeek has released two preview versions of its new large language model, DeepSeek V4, in Flash and Pro variants. Both are mixture-of-experts models with 1 million token context windows. The Pro model is the largest open-weight model available at 1.6 trillion total parameters (49B active), surpassing competitors like Kimi K 2.6 and MiniMax M1. DeepSeek claims the models nearly close the gap with frontier models on reasoning benchmarks and outperform GPT-5.2 and Gemini 3.0 Pro on some tasks, though they trail state-of-the-art models by roughly 3–6 months on knowledge tests. Pricing is significantly lower than competing frontier models. Both models are text-only and lack multimodal capabilities.