ExpectedDeepSeek
DeepSeek V4-Pro Weights
Open-weight foundation model packaging in final preparations following initial API serving evaluation benchmarks.
2026 · Q4✓ Sourced
DeepSeek
DeepSeek-V
Explore upcoming DeepSeek-V foundation models, high-efficiency mixture-of-experts architectures and open releases.
Official website DeepSeekThe DeepSeek-V Architecture
The DeepSeek-V family serves as DeepSeek’s high-efficiency foundation base, pioneering Multi-head Latent Attention (MLA) and DeepSeekMoE architectures.
Training Efficiency and Scale
With low active parameter counts per token and innovative load balancing, DeepSeek-V delivers state-of-the-art capability at unprecedented compute efficiency.
Verifiable Open Releases
We monitor planned open-weight drops and API serving updates, backed by authenticated developer repository commits.
Next on the horizon · this page
ExpectedDocumented releases
Timeline · this page
DeepSeek V4-Pro Weights — Open-weight milestone anticipated following initial API serving benchmark evaluations.