Key highlights
- DeepSeek-V4 preview was open-sourced and has 1.6 trillion parameters.
- Sinopec deployed the DeepSeek-V4-Flash model into its private domestic intelligent computing resource pool on the model's release day and integrated it into the Great Wall AI Platform.
- The model is optimized for multiple domestic computing chips, and Sinopec's team completed deployment, performance testing and optimization after obtaining the weights.
Model release
On April 24 DeepSeek released and open-sourced the preview version of DeepSeek-V4, a new architecture with 1.6 trillion parameters.
Private deployment
Sinopec deployed the DeepSeek-V4-Flash model into its private domestic intelligent computing resource pool on the model's release day and integrated it into the Great Wall Artificial Intelligence Platform to serve the group.
Architecture and optimization
DeepSeek-V4 is designed to match top closed-source models in knowledge and reasoning performance and is adapted for multiple domestic computing chips; Sinopec's technical team completed deployment, performance testing and optimization after obtaining the open-source model weights.
