# DeepSeek
Latest news and articles about DeepSeek
Total: 89 articles found
Beyond Reasoning: Why Agentic Thinking Is the New Frontier for Global AI
Former Alibaba Qwen lead Lin Junyang argues that AI is shifting from a 'reasoning' phase to an 'agentic' phase where models are trained to prioritize action and environmental interaction. He highlights the technical difficulties in merging deep thinking with instruction-following and predicts that future AI success will depend on building integrated systems that can independently determine the necessary level of deliberation for any given task.
Alibaba’s Open-Source Gambit: How the RISC-V ‘Xuantie’ Aims to Disrupt the Global Chip Duopoly
Alibaba has launched the record-breaking Xuantie C950 RISC-V CPU, signaling a strategic shift toward high-performance, open-source AI computing. With its chip unit T-Head achieving mass production and NVIDIA's CUDA adding support, the RISC-V architecture is emerging as a credible third pillar to challenge the x86 and ARM duopoly in the global semiconductor market.
Digital Bridges: China and Singapore Forge a New Frontier in AI Cooperation
At a recent dialogue in Singapore, officials and industry leaders highlighted the rapid integration of Chinese AI models like DeepSeek into the Singaporean ecosystem. The partnership between Singapore's national 'SEA-LION' model and Alibaba Cloud marks a significant milestone in regional technological synergy.
DeepSeek’s DualPath Promises to Halve AI Inference Costs — But Questions Remain
DeepSeek has introduced DualPath, an inference architecture it says can double efficiency and lower the compute cost of running large AI models. The move reflects a broader industry shift toward software and architectural optimisations that could reduce reliance on cutting‑edge chips, but real‑world validation and integration challenges remain.
Musk’s Bold Claim Fuels China’s New Year Rush to Build AI That Writes Code
Elon Musk’s claim that AI may soon eliminate the need for human programmers has sharpened an already heated competition in China, where major firms have released coding‑focused models during the Spring Festival. The combined effect of domestic model improvements, falling tool prices and early commercial traction promises big productivity gains but also raises reliability, security and labour‑market challenges.
Seedance 2.0 and the Moment AI Video Became Industrially Real
ByteDance’s Seedance 2.0 has turned AI video generation into a commercially viable technology, producing near‑cinematic output quickly and cheaply. The model has intensified competition between Chinese and international labs, accelerated industry moves toward AI‑led content production, and raised urgent questions about IP, regulation and platform power.
Chinese AI Lab DeepSeek Trials 1‑Million‑Token Context Window in App — API Still Capped at 128K
DeepSeek is testing a new long‑context model in its web and app interfaces that supports roughly one million tokens, while its public API remains limited to 128K token context on version 3.2. The trial highlights the commercial and technical trade‑offs involved in bringing ultra‑long context windows to production and signals intensifying competition in China’s AI landscape.
China’s ¥100bn Lunar New Year ‘Red‑Packet’ War Exposes an Emerging AI Compute Crunch
China’s Spring Festival promotional campaigns—collectively worth nearly ¥100 billion—have driven extreme traffic spikes that briefly knocked services offline and highlighted a growing mismatch between consumer‑facing AI adoption and available compute capacity. Firms are ramping cloud and data‑centre investments even as advances in models and token windows multiply inference demand, creating cross‑cutting pressures on energy systems and commodity supply chains.
China’s DeepSeek Pushes Context Limits — and Triggers a Backlash Over a Colder, ‘Faster’ Model
DeepSeek activated a grayscale update extending context length to 1 million tokens, prompting user complaints that the assistant sounds colder and less personalised. Industry sources say the build is a speed‑focused variant intended to stress‑test long‑context performance ahead of a V4 launch, highlighting trade‑offs between throughput and conversational quality. The episode illustrates the wider tension in scaling LLMs: architectural gains can come at the cost of user experience and trust.
A Night of Acceleration: Zhipu’s GLM‑5 Debuts as MiniMax and DeepSeek Race to Keep Up
Three leading Chinese AI firms unveiled near‑simultaneous upgrades that signal a shift from demo‑level coding assistants to production‑oriented, agentic systems. Zhipu launched GLM‑5 as an open‑source foundation for long‑horizon engineering tasks, while MiniMax and DeepSeek pushed product and context upgrades aimed at real‑world throughput and extended interactions.
DeepSeek's Quiet Leap: 1‑Million‑Token Context and May‑2025 Knowledge Cut Hint at a Next‑Gen Chinese LLM
DeepSeek has begun limited testing of a model that supports a 1 million token context window and uses training data up to May 2025, a significant expansion from its previous 128k limit. The change suggests material architectural or pipeline upgrades and signals intensified competition among Chinese AI providers to ship more capable, enterprise‑ready models.
WeChat Clips Tencent’s Yuanbao in China’s AI ‘Red‑Envelope’ War — A Lesson in Platform Governance
WeChat has blocked in‑chat links from Tencent’s AI app Yuanbao for using share mechanics that the platform said induced excessive forwarding and harmed user experience, forcing Yuanbao to change its sharing approach. The enforcement, which also affected Baidu and Alibaba apps, underscores how platform governance and ecosystem fit now matter as much as model performance or marketing spend in China’s heated AI ‘red‑envelope’ competition.