[DeepSeek Releases V3.1 Model, Supporting Proxy and Hybrid Reasoning] DeepSeek has announced the launch of the V3.1 model, introducing the “Think & Non-Think” hybrid reasoning mode for the first time. Users can switch modes via the “DeepThink” button. The reasoning speed of the V3.1-Think mode has been improved compared to DeepSeek-R1-0528, with enhanced tool usage and multi-step proxy task capabilities. The API now supports 128K context, is compatible with the Anthropic API format, and adheres to strict function calls. The V3.1 base model extends long-context capabilities based on 840B tokens, with open-source weights already available. Pricing adjustments will take effect on September 5.