DeepSeek has released a preview version of its V4 model, claiming top-tier reasoning performance and significantly...

DeepSeek has released a preview version of its V4 model, claiming top-tier reasoning performance and significantly improved agent capabilities. The model is now available on web, app, and API source. This launch indicates ongoing investment in model scaling and performance, which could sustain demand for AI training and inference hardware. Separately, a Manifold Markets prediction market shows only 4.01% belief that DeepSeek lied about the GPU compute budget for V3 source, suggesting the market largely trusts DeepSeek's reported efficiency. Combined, these signals point to continued AI model competition and potential upward pressure on GPU procurement by AI labs, though the evidence is based on a single product announcement and a low-probability rumor. Therefore, the signal is low confidence.