DeepSeek updates its AI model with V3.2
DeepSeek has introduced its latest ‘experimental’ AI model, V3.2, which is an intermediate step towards its next-generation architecture. The model features ‘DeepSeek Sparse Attention’ (DSA), enhancing long context performance while reducing compute costs. Despite some marginal performance drawbacks compared to the previous V3.1, the model is seen as a significant development. DeepSeek is cutting API prices by more than 50% to support this release. The new architecture is expected to be a major product release since the V3 and R1 models.
Source: https://www.siliconrepublic.com/machines/anthropic-deepseek-ai-model-sonnet-4-5-claude-v3-2-exp-china Silicon Republic