DeepSeek releases the DeepSeek V4 preview
DeepSeek announced V4 Pro and V4 Flash preview models with open weights, long-context support, and updated API access.
The official DeepSeek release page introduced V4 Pro and V4 Flash, linked open weights and a technical report, and described a one-million-token context configuration.
DeepSeek also documented API model names and a retirement date for older aliases. AI3Radar records the original release state and does not treat later availability as implied by this historical item.