AI News · 2026-04-24

DeepSeek V4 preview: 1.6T parameter open-weight model closes gap with frontier

DeepSeek launched two preview versions of DeepSeek V4 — V4 Pro with 1.6 trillion total parameters (49B active) and V4 Flash with 284 billion parameters (13B active) — both using mixture-of-experts architecture with 1M token context windows. The models outperform GPT-5.2 and Gemini 3.0 Pro on reasoning benchmarks and achieve performance comparable to GPT-5.4 on coding competition tests, while trailing frontier models by an estimated 3 to 6 months.