Open-source releaseModerate global significanceConfirmed confidence
DeepSeek releases the efficiency-focused DeepSeek-V2
DeepSeek releases a 236B mixture-of-experts model that combines sparse activation with latent attention to cut training and inference costs.
What happened
Event details
DeepSeek-V2 brought two of the lab's core efficiency ideas into one open-weight model: the DeepSeekMoE expert design and Multi-head Latent Attention, which compresses the key-value cache used during inference. The model has 236 billion total parameters with roughly 21 billion active per token. DeepSeek's report framed the release around lower training cost, much smaller KV-cache requirements, and higher serving throughput rather than parameter count alone.
Assessment
Why it matters
V2 proved that DeepSeek's efficiency work was a deployable architecture, creating the foundation later scaled into V3.
67/100Global significance score. Regional effects are recorded only when the evidence supports a meaningful difference.