[DeepSeek Releases Visual Primitives Reasoning Method to Enhance Multimodal Complex Reasoning Capabilities] According to the technical report released by DeepSeek, the company has proposed the "Visual Primitives" reasoning method, which embeds basic visual units such as points and boxes into the reasoning chain to address the Reference Gap issue in multimodal tasks. This method is based on the DeepSeek-V4-Flash architecture and achieves low image token consumption by compressing KV cache. In benchmarks for counting and spatial reasoning, its performance is comparable to GPT-5.4, Claude-Sonnet-4.6, and Gemini-3-Flash (limited to certain dimensions). The team stated that they plan to open-source some benchmarks and data in the future, and the model weights will be integrated and released subsequently.