According to Reuters, DeepSeek’s latest model V3.2-Exp introduces a new sparse attention mechanism, enabling more efficient handling of long sequences while reducing computational overhead.
The company also announced API price cuts of 50%+, positioning this release as a strategic move to pressure rivals both domestically (e.g. Alibaba’s Qwen) and internationally (e.g. OpenAI).
DeepSeek’s ambition is to use this intermediate model as a bridge until its next-generation architecture is ready, hoping that performance gains and cost reductions will attract adoption and momentum.
Key Highlights
DeepSeek releases V3.2-Exp, an “intermediate” model with sparse attention innovation
Applies 50%+ API pricing cuts to enhance competitiveness
The model is positioned as a precursor to next-gen architectures
Aims to push pressure on both Chinese and global AI rivals
Why This Matters
Efficiency & cost breakthroughs: Innovations like sparse attention can reduce barriers for training and inference at scale.
Aggressive pricing strategy: By cutting API costs deeply, DeepSeek aims to grow market share rapidly.
Strategic positioning: Framing it as an intermediate model allows DeepSeek to iterate quickly and compete continuously.
Global AI pressure: Moves from Chinese AI firms like DeepSeek will be watched closely by both domestic and international competitors.
Source
Reuters – Full Article
Stay up-to-date
