DeepSeek(@deepseek_ai):🧠 Asymmetric architecture. More intelligence, less cost. 🔹 552B-parameter MoE. 🔹 New Causal Encoder–Decoder architecture: just 8B active parameters for input, 16B for output. 🔹 New pre-training methods + larger-scale RL post-training deliver benchmark results ahead of flagship models, including DeepSeek-V4-Pro. 2/6

DeepSeek

@deepseek_ai

Unravel the mystery of AGI with curiosity. Answer the essential question with long-termism.

Joined October 2023

0 Following 1.1M Followers

DeepSeek@deepseek_ai

2026.09.10 06:10

🧠 Asymmetric architecture. More intelligence, less cost. 🔹 552B-parameter MoE. 🔹 New Causal Encoder–Decoder architecture: just 8B active parameters for input, 16B for output. 🔹 New pre-training methods + larger-scale RL post-training deliver benchmark results ahead of flagship models, including DeepSeek-V4-Pro. 2/6