ENLatest AI NewsTencent Hunyuan Launches Hy ASR 3.0 Preview: Speech Recognition No Longer Just Converts Words, but Truly Understands the ContextTencent Hunyuan launches the speech recognition model Hy ASR 3.0 preview, which leverages the language comprehension capabilities of the large language model Hy3, combining high-precision recognition with deep semantic understanding. It has been significantly improved in general recognition, context awareness, multi-scenario robustness, and dialect coverage. It can provide more accurate and coherent transcriptions by combining context, achieving an evolution from word-by-word transcription to semantic understanding.6 hours ago55.1KAI Writing Code Burns 12.15 Million Yuan, Exceeding Budget by 860%: Amazon Internal Incident Sounds the Alarm for Corporate AI CostsU.S. media exposed an AI cost control incident at Amazon: Engineers used Anthropic’s Claude Sonnet model to handle routine tasks such as matching authors with product data, resulting in an $18 million bill, exceeding the budget by 860%. The project was delayed for five months and ultimately failed to be implemented, leading to massive waste due to high computing power consumption.6 hours ago13.4KTencent Hunyuan Launches Hy ASR3.0 Preview: Speech Recognition Enters a New Era of Understanding ContextTencent Hunyuan launches Hy ASR3.0preview, integrating Hy3 LLM’s semantic understanding into speech recognition for context-aware, direct transcription. On open benchmarks, word error rates reach 3.34% for Mandarin, 2.62% for English, and 3.12% for Cantonese, achieving around 3% overall with impressive multi-dimensional performance…..7 hours ago14.1K8B Parameters Integrated into 4K Generation and Editing, SenseTime Opens Source Lightweight Unified Multimodal Model U1.5-Lite Preview VersionSenseNova U1.5-Lite-Preview, an open-source unified multimodal lightweight model by SenseTime, systematically iterates on U1 architecture, integrating visual understanding, reasoning, generation, and editing. With only 8B-MoE parameters, it supports 4K resolution, finer textures, and complex visual control, advancing multimodal capabilities in a compact size…..7 hours ago14.2KMegMind Open Sources ForgeStencil: Dual Agent Automatically Optimizes Over 100 Industrial Software in One Week with Zero Human InterventionStencil computation, a core pattern in modern industry and advanced research, iteratively updates grid points using weighted neighbor combinations. It is widely applied in weather forecasting, seismic exploration, and electromagnetic simulation. Extremely memory-intensive, its speed is limited by memory bandwidth, dominating runtime and directly determining overall performance…..8 hours ago13.0KSwiftlet fits an 80B-level Qwen giant into Mac: peak memory is only 4.3GB, iPhone 17 can also run 35B nativelyThe Swiftlet project uses Swift + Metal to enable MoE models like Qwen3-Next to run with minimal size on Mac: only a small dense core is kept in memory, and expert weights are stored on SSD and loaded on-demand. On an M5 Mac, the 4-bit version of Qwen3.6-35B-A3B occupies only 18GB of disk space, with a peak memory of 2.6GB and a decoding speed of 7–11 tok/s.8 hours ago33.2KSenseTime Releases 8B Light Cannon U1.5-Lite-Preview: Not Big in Parameters, But Drives Closed-Source Models CrazySenseTime releases the open-source lightweight multimodal model SenseNova U1.5-Lite-Preview, based on the original NEO-Unify architecture. With only 8B-MoT parameters, it achieves generation and editing quality comparable to closed-source models. It supports multimodal understanding, reasoning, and 4K high resolution, challenging the large model arms race with its compact size.8 hours ago12.6KCodeBuddy and WorkBuddy Extend the Free Trial of Hy3 Model Until August 31stCodeBuddy and WorkBuddy have extended the free access period for the Hy3 large language model to August 31, 2026, due to high user demand. Hy3 supports text-only tasks with no multimodal capabilities. If users attempt image or video generation via these products, the system will automatically switch to appropriate multimodal models and deduct credits as normal…..8 hours ago11.3KApple Urges Injunction to Block OpenAI: Former Executive Takes Trade Secrets to the iPhone Throne May Be ThreatenedApple filed a preliminary injunction on Monday to prevent two former employees who joined OpenAI and OpenAI from using or disclosing alleged trade secrets, warning of irreparable harm without the ban. The dispute stems from Apple’s lawsuit last month accusing OpenAI and the two ex-employees of misappropriating secrets to advance OpenAI’s consumer AI products…..8 hours ago11.9KBeijing Bar Launches Token Free Drinking: Get Free DeepSeek V4 Flash Computing Power by Connecting to Wi-FiAGI Bar in Beijing has launched a ‘Token Free Drinking’ service, allowing customers to freely and unlimitedly access the DeepSeek V4 Flash model API after connecting to the store’s Wi-Fi. The bar is eq