AI Models
Explore available models and their capabilities
DeepSeek
Released December 2025
DeepSeek V3.2
Efficient reasoning, sparse attention, math olympiad
Reasoning
Tool Use
About this model
DeepSeek V3.2 balances computational efficiency with strong reasoning capabilities. It introduces DeepSeek Sparse Attention (DSA), a fine-grained sparse attention mechanism that reduces training and inference cost while maintaining quality for extended contexts. The system uses reinforcement learning post-training to enhance reasoning performance and demonstrates competitive results on mathematical olympiad problems.
Context Window
163K
Input Cost
$0.25/M
Output Cost
$0.38/M
Input Types
Text