DeepSeek V4 Pro Launches With 1.6 Trillion Parameters

DeepSeek released V4 Pro on April 24, a mixture-of-experts model with 1.6 trillion total parameters — though only 49 billion activate per inference — and a verified 1-million-token context window, under an MIT open-source license at notably low pricing. The combination of genuine long-context capability, open weights, and pricing well below comparable Western models at $1.74 per million input tokens positions it as a serious competitive threat to Anthropic, OpenAI, and Google in the agentic and enterprise workload market. Whether Chinese export controls or geopolitical pressure leads to restrictions on Western use of DeepSeek infrastructure remains the key watch item, as the model is already running in production deployments via NVIDIA NIM. The model ships with dual Think / Non-Think reasoning modes.

Sources