DeepSeek V4 Pro

About this model

DeepSeek V4 Pro is a 1.6 T parameter Mixture-of-Experts (MoE) chat model from DeepSeek AI. It features a hybrid attention architecture with compressed sparse and heavily compressed attention, supporting a context window of one million tokens. The model achieves 90.1 % EM on MMLU, 76.8 % Pass@1 on HumanEval, and 51.5 % EM on LongBench‑V2, demonstrating strong language, coding, and long‑context capabilities. It is released under the MIT License.

Technical specifications

Capabilities

Input modalities

Output modalities

Reasoning

Hybrid Default on

Knowledge horizon