Full Summary
This Monday morning, DeepSeek has permanently slashed the price of its V4-Pro AI model by a remarkable 75%, a move confirmed by both the *Latest News from Azerbaijan* and *WinBuzzer*. Developers using DeepSeek’s API will now pay between 0.025 and 6 yuan per million tokens, a substantial drop that could significantly reduce costs for businesses relying on high-volume AI services. Meanwhile, Google is rolling out its new AI video model, Gemini Omni Flash, in the Gemini app, Google Flow, and YouTube Shorts. *Pune Mirror* reports this tool creates short, editable video clips from various inputs like text or images, with every video carrying an imperceptible SynthID watermark for verification. *Gotrade* adds that this is part of Alphabet's new agentic AI lineup, featuring the cost-competitive Gemini Flash 3.5 model, priced at roughly half of competing models. In other AI news, xAI's Grok foundation model V9-Medium has finished training and is expected for public release in two to three weeks, according to *BASENOR - Tesla Accessories*. This 1.5 trillion parameter model, three times larger than its predecessor, was trained with significant Cursor data, suggesting a focus on developer and coding applications. *News-Medical* announces a novel AI model called resLens, designed to improve the detection of antibiotic resistance genes, while *the-decoder.com* reveals that AI models often provide correct answers but cite the wrong sources, a problem dubbed "attribution hallucination." This issue, highlighted by the new CiteVQA benchmark, saw GPT-5.4's performance drop from 87.1 for answer quality to just 59 when correct sourcing was needed. Finally, the METR AI benchmark, a popular graph central to discussions about AI's capabilities, is facing renewed scrutiny. *Startup Fortune* reports a debate questioning its underlying methodology, suggesting that while the graph appears clear, it may be complex and potentially misleading, influencing investors and policymakers on AI's impact. This means the widespread understanding of AI's capabilities, and the investments made based on it, could be built on shaky ground, potentially affecting future job markets and technological advancements.