Sinopec Boosts Russian ESPO Oil Purchases as Middle East Supply Tightens 
Nvidia Seen Beating Q2 Targets as Vera Rubin Cycle Begins 
Apple Restores Telegram to App Store After Content Policy Violation 
Airbnb Stock Jumps After Q2 Earnings Beat, Strong 2026 Outlook 
Meta Ordered to Pay $567M Over Child Safety Violations in New Mexico 
Alphabet Stock Slides as Google AI Pioneer Jeff Dean Exits to Launch Discovery Loop 
Nvidia to Invest Up to $3 Billion in Blackstone-Backed Lancium 
Shein Scales Back Vietnam Operations as US Trade Rules Shift 
SK Hynix, Samsung Lead Asian Chip Stock Selloff After Sandisk, Western Digital Outlook 
Qantas Shares Climb as Long-Haul Pilot Deal Eases Strike Concerns 
Samsung, SK Hynix Test AMEC Chipmaking Tools for China Backup Plan 
Western Digital Q4 Earnings Beat Estimates as FY2027 Outlook Tops Expectations 
SpaceX Earnings Preview: Bernstein Says 4 Key Factors Will Drive Long-Term Valuation 
SK Hynix Bonus Dispute Deepens as Union Rejects Stock-Based Payout Proposal 
DeepSeek to Raise AI API Prices as Demand for New Models Surges 
Treasury Wine Estates Shares Jump as U.S. Overhaul Lifts FY2026 Outlook 
DeepSeek has unveiled preview versions of its latest open-source artificial intelligence models, marking a significant step forward in AI performance and reasoning capabilities. The new releases, DeepSeek-V4-Pro and DeepSeek-V4-Flash, are designed to compete with leading global AI systems while maintaining efficiency and scalability.
The DeepSeek-V4-Pro model stands out with an impressive 1.6 trillion parameters, positioning it among the most powerful open-source AI models available today. Meanwhile, the DeepSeek-V4-Flash offers a more lightweight alternative with 284 billion parameters, making it suitable for applications that require faster processing and lower computational costs. Both models feature an extended context window of up to 1 million tokens, allowing them to process and retain significantly larger amounts of information compared to earlier AI systems.
Currently, the models are limited to text-based processing, but DeepSeek has confirmed ongoing development toward multimodal capabilities, which will enable support for images and video in future updates. This move aligns with broader industry trends as AI companies race to build more versatile and adaptive systems.
According to DeepSeek, its upcoming V4-Pro-Max model delivers top-tier results in coding benchmarks and narrows the performance gap with leading closed-source AI models in reasoning and agent-based tasks. On the MMLU-Pro benchmark, DeepSeek-V4-Pro reportedly performs on par with OpenAI’s GPT-5.4, while slightly trailing competitors such as Google’s Gemini-3.1-Pro and Anthropic’s Claude Opus 4.6.
While DeepSeek has not disclosed the exact hardware used for training, reports suggest a shift toward Huawei’s advanced AI chips, particularly the Ascend 950, which now fully supports the V4 models. This comes amid earlier scrutiny over potential use of restricted NVIDIA hardware.
The release marks DeepSeek’s first major AI model launch since its R1 model disrupted the industry in early 2025 by delivering high performance with greater efficiency. The company continues to gain momentum, with recent reports indicating that major Chinese tech firms like Tencent and Alibaba are considering investments that could value DeepSeek at over $20 billion. As one of China’s leading AI startups, DeepSeek is rapidly strengthening its position in the global artificial intelligence landscape.
