AIAhead of AI7 months ago
DeepSeek V3.2: Architecture and RL Updates
30 min read
DeepSeek released V3.2, an open-weight flagship model with performance matching GPT-5 and Gemini 3.0 Pro. The update introduces sparse attention and reinforcement learning changes. V3.2 builds on the V3/R1 architecture from December 2024, which gained popularity through the R1 reasoning model.
Level
Hype check
Tap to vote and see what everyone thinks.
