According to DeepSeek (@deepseek_ai), DeepSeek has updated its API with significant enhancements for enterprise AI development. The deepseek-chat API now supports 'non-thinking mode,' while deepseek-reasoner introduces 'thinking mode,' catering to different AI application needs. Both APIs now feature a 128K context window, enabling advanced large-context processing for complex tasks. Additionally, the APIs support the Anthropic API format, which increases compatibility for developers migrating from Claude or other Anthropic-based systems. The Beta API also offers strict function calling, streamlining workflow automation and task orchestration in business applications. These updates provide more robust API resources, smoother performance, and open new opportunities for building large-scale, reliable AI solutions across industries (Source: DeepSeek Twitter, August 21, 2025).

According to DeepSeek (@deepseek_ai), the latest upgrades to their AI tools and agents have delivered significantly better results on SWE and Terminal-Bench benchmarks, highlighting stronger multi-step reasoning for complex search tasks and substantial gains in thinking efficiency. These technical improvements are particularly relevant for AI-powered developer tools, coding assistants, and enterprise search solutions, where robust reasoning and efficient task execution drive productivity and business value. (Source: DeepSeek Twitter, August 21, 2025)

According to DeepSeek (@deepseek_ai) on Twitter, DeepSeek AI has announced significant pricing changes for its API services. New pricing rates will take effect and current off-peak discounts will end on September 5th, 2025, at 16:00 UTC. Until that date, API usage will continue to follow the current pricing structure. This change is likely to impact businesses integrating DeepSeek's AI APIs into their products, especially those optimizing for cost during off-peak hours. For detailed information on upcoming pricing, users are directed to the official pricing page. Source: DeepSeek (@deepseek_ai), August 21, 2025.

According to DeepSeek (@deepseek_ai), the new DeepSeek-V3.1 AI model introduces a hybrid inference system, allowing both 'Think' and 'Non-Think' modes in a single model. This upgrade enables faster reasoning, as DeepSeek-V3.1-Think significantly outpaces the previous DeepSeek-R1-0528 in delivering answers. Additionally, post-training enhancements empower the model with stronger agent skills, improving its ability to utilize tools and execute complex tasks autonomously. These advancements signal a move toward more capable AI agents, opening up business opportunities in process automation, intelligent workflow management, and advanced digital assistants (Source: DeepSeek on Twitter, August 21, 2025).

According to DeepSeek (@deepseek_ai), the company has released the V3.1 Base model, which features continued pretraining on 840 billion tokens for improved long context extension. The update also includes an overhauled tokenizer and chat template, aiming to enhance language model performance for extended conversations. Both the V3.1 Base and full V3.1 model weights have been open-sourced, offering developers and AI businesses access to advanced large language model capabilities. This release marks a significant step in open-source AI development, enabling enterprises to deploy long-context chatbots and advanced NLP applications with greater efficiency and scalability (Source: DeepSeek Twitter, August 21, 2025).

According to DeepSeek (@deepseek_ai), the newly released DeepSeek-R1-0528 introduces significant upgrades including improved benchmark performance, enhanced front-end capabilities, and a notable reduction in AI hallucinations. The update also adds support for JSON output and function calling, allowing for greater integration into business workflows and improved reliability in enterprise applications. API usage remains unchanged, ensuring seamless adoption for existing developers. These advancements present notable opportunities for businesses seeking robust, production-ready AI solutions with increased accuracy and integration flexibility (source: DeepSeek on Twitter, May 29, 2025).