AI NEWSLatest AI NewsArticleTencent Hunyuan Hy4 Preview Open Source: 770B Parameters, Million-Context, Standing at the Forefront of Open SourcePublished in Latest AI NewsTime :Aug 28, 2026Read :4minuteTencent has released the Hy4 preview model today, with a total of 770B parameters and 49B activated parameters, and a context length of up to 1M. The official stated that the model has significantly expanded in size, context, and data scale, and the simultaneous advancement of pre-training and post-training has led to another leap in intelligence level, firmly securing its position among the top open-source models. The model is now open-sourced and has been synchronized on HuggingFace, GitHub, ModelScope, and Gitcode, and has also been integrated into Tencent Cloud TokenHub and OpenRouter.Productivity scenarios have been fully upgraded, outperforming GLM-5.3 and Kimi K3 in blind testsBased on high-quality data co-built with experts in software engineering, gaming, finance, and security within Tencent, Hy4 preview has made significant progress in real tasks: it strengthens long-term development understanding, planning, and verification on the software engineering side; it can complete full delivery from data processing to document table presentation in office analysis; it can generate playable prototypes with one requirement in game development; and it has made progress in research scenarios such as molecular dynamics and condensed matter physics.Tencent organized 163 internal experts and 203 engineering tasks for blind testing, and Hy4 preview achieved an average score of 2.99 (out of 4.00), slightly surpassing GLM-5.3’s 2.92 and Kimi K3’s 2.94. Along with Hyra, it has also made breakthroughs in the century-old geometric problem of three-dimensional Blaschke–Lebesgue, advancing the volume lower bound from 0.380799 to 0.41104, with only about a 2% gap remaining to the Meissner tetrahedron conjecture of 0.41986.The official admitted that Hy4 preview is just an early iteration of Hy4, and there is still room for improvement in pre-training and post-training. It also has known issues such as complex task long thinking and excessive self-verification, and will continue to iterate agilely. The model has been deeply integrated with CodeBuddy and WorkBuddy, and users can also directly experience it in products like Yuanbao and ima.

AI NEWSLatest AI NewsArticleTencent WorkBuddy Launches Integration with Hy4preview: Domestic and Overseas Versions Synchronized, Two-Week Free Trial StartedPublished in Latest AI NewsTime :Aug 28, 2026Read :3minuteTencent WorkBuddy announced on August 28 that it has launched the new open-source large model Hy4preview of Tencent Huan Yuan. The domestic and international versions were launched simultaneously, and a limited-time free policy was introduced: Hy4preview offers a two-week free trial, and the free usage period for the previous generation Hy3 has been extended by one month to September 30th.Hy4preview is the official open-source flagship model released by Tencent Huan Yuan. It has a total of 770B parameters and 49B activated parameters, with a context length of up to 1M. It is positioned for real productivity scenarios such as coding, office work, and science. It has been open-sourced on platforms such as HuggingFace, and is also available on Tencent Cloud TokenHub and OpenRouter.The domestic and international versions of WorkBuddy were launched simultaneously, allowing developers to directly call this model in programming and intelligent office workflows, achieving rapid integration from model release to product implementation.This combination strategy of initial access and limited-time free offers aims to attract developers to test the model’s capabilities and collect real feedback with zero barriers, while extending the free period of Hy3 to retain existing users. This continues Tencent Huan Yuan’s approach of promoting the accessibility of large models through open source and product ecosystem collaboration. The focus of competition among model manufacturers is shifting from parameter scale to productization and ecosystem service capabilities.

AI NEWSLatest AI NewsArticleOpenClaw 2.0 Released: Reconstructed the Infrastructure, Supports Multi-User Shared Cloud SessionsPublished in Latest AI NewsTime :Aug 31, 2026Read :3minuteOn August 30, the open source AI project OpenClaw released its largest update to date, OpenClaw 2.0. This update was completed by 933 contributors, with over 16,000 pull requests merged, covering about half of the code submissions in the project’s history. Due to the development workload exceeding the capacity of the original infrastructure, the team spent nearly seven weeks rebuilding the product foundation and delivery process.OpenClaw 2.0 focuses on simplifying installation and reworking the browser experience, and then carries out a deep overhaul of the entire process including messages, memory, skills, models, automation, and security mechanisms. The new version greatly reduces configuration steps, prioritizing the use of existing local API keys, models, or ChatGPT/Claude subscriptions, achieving an extremely simple out-of-the-box experience, with more complex configurations completed through interaction later on.For daily scenarios, the system supports cross-platform connectivity and automated task processing (such as automatically parsing emails and pushing notifications, retrieving information across applications and messages, etc.). In addition, the new version introduces the “Shared Cloud Session” feature, breaking the previous context isolation of individual intelligent agents, allowing team members to collaborate in real-time or fully hand over the AI processing flow.As an open-source project, OpenClaw 2.0 not only improves the deployment efficiency of intelligent agents, but also marks the evolution of AI intelligent agents from single-machine auxiliary tools to collaborative and autonomous productivity infrastructure.

AI NEWSLatest AI NewsArticleBaidu Brings in North American Large Model Expert, Former DeepSeek Researcher Now Head of Wenxin MultimodalPublished in Latest AI NewsTime :Sep 3, 2026Read :3minuteThe Baidu Foundation Model Research Department (BMU) has recently completed a key talent deployment. On September 3rd, Sun Tianxiang, the head of BMU, announced internally that he had recruited a senior researcher Wu Qin (alias) from a Frontier Lab in North America, to strengthen the pre-training capabilities of the ERNIE large model. Wu Qin has participated in the development of multiple generations of large models over the past two years, but his specific background remains undisclosed.At the same time, Wei Haoran, a former researcher at DeepSeek, officially transferred to BMU and now serves as the head of the multimodal algorithm team for ERNIE. Wei Haoran joined Baidu in early 2026. His team released the Unlimited OCR model in June this year, achieving the top global score in the OmniDocBench benchmark test, setting a new record for end-to-end OCR performance. Previously, he led the development of versions V3.2 and V4 at DeepSeek, as well as the DeepSeek-OCR series of open-source models, and has extensive experience in developing multimodal large models.It is reported that Sun Tianxiang, born in 1997, took office as the head of BMU in July this year and joined the Baidu Model Committee and the Technical Strategy Committee. With the core team in place, he is driving a dual upgrade of the ERNIE large model in both organizational and technical aspects, further aligning with the international frontiers of model development by introducing top talents.

AI NEWSLatest AI NewsArticleAnthropic Releases Claude Fable5.1: Enhanced Agent Coding Capabilities, Up to 45% Cost ReductionPublished in Latest AI NewsTime :Sep 3, 2026Read :3minuteAnthropic officially announced the launch of its new generation AI modelsClaude Fable5.1and the restricted variantMythos5.1, significantly enhancing agent coding and scientific reasoning capabilities while substantially optimizing inference costs.Both models are based on the same foundation. Fable5.1 fully opens its API to public cloud platforms, while Mythos5.1 is restricted to specific projects in cybersecurity and life sciences. At the same time, both new models are the first Claude products to have built-in AI text watermarks.In terms of performance, Fable5.1 demonstrates strong agentic task processing capabilities, achieving a score of 55.8% in the Terminal-Bench4.0 agent coding test (Mythos5.1 scored 60.9%), and 52.6% in the Terminal-Bench-Science0.1 scientific research test. This represents a doubling compared to the previous version Fable5 and surpasses competitors like GPT-5.6Sol.Additionally, the new models have optimized the safety filtering mechanism, reducing the false positive rate in cybersecurity by 60%, and the false positive rate for biologically harmless queries by 85%, enabling them to autonomously identify software vulnerabilities.Economic efficiency was the core focus of this upgrade. By significantly lowering the cost of cache reads to $0.25 per million tokens, Fable5.1 can save approximately 25% in costs for typical workloads, and up to 45% for complex agent workflows running over extended periods.

AI NEWSLatest AI NewsArticleAI Performance Evaluation Shifts from Usage to Output: Meta Abandons Adoption Dashboards and Token MetricsPublished in Latest AI NewsTime :Sep 3, 2026Read :3minuteMeta engineering executives Maher Saba and Santosh Janardhan released a memo to internal employees on Tuesday, updating the company’s expectations for engineers, explicitly stating that “the company will not use AI adoption dashboards or token usage to evaluate employee impact.” The memo shows that 93% of code changes at Meta are now assisted by AI agents, and AI is deeply embedded in daily engineering processes. However, the document also emphasizes that adopting AI “is not the ultimate goal itself.“This statement points to a common practice in the tech industry over the past two years: companies have generally included metrics such as code generation ratio, Copilot adoption rate, and token consumption in internal dashboards to promote AI tool implementation. Some companies further link these metrics to performance evaluations. This type of quantitative management has accelerated tool penetration but has also led to “performative AI use” focused on quantity.By setting boundaries, Meta effectively separates tool usage intensity from actual output, returning the evaluation of employee impact back to the delivery results themselves.From an industry perspective, Meta’s 93% penetration rate indicates that AI programming agents have reached an infrastructure level in large company engineering systems. At this point, continuing to evaluate “whether it is used” has no incremental value. Shifting the assessment focus from adoption rate to output quality aligns with the overall trend of enterprise management logic moving from the promotion phase to the practical results phase after AI programming tools have been widely adopted.

AI NEWSLatest AI NewsArticleBroadcom predicts Anthropic and OpenAI will surpass Google to become the top two customers for AI custom chipsPublished in Latest AI NewsTime :Sep 3, 2026Read :4minuteDuring the third quarter of fiscal year 2026, Broadcom’s President Chen Fuyang made a major forecast during the company’s earnings call, stating that as competition for computing infrastructure among large AI model companies intensifies, Anthropic and OpenAI are expected to surpass Google to become the two largest core customers in Broadcom’s AI-specific integrated circuit (ASIC) business. Specifically, Anthropic is expected to become Broadcom’s largest XPU custom chip customer between 2027 and 2028, while OpenAI is anticipated to follow closely as the second-largest customer.In the just-completed third quarter, Broadcom’s AI semiconductor business showed strong growth, with revenue reaching $16.7 billion, representing a significant increase of 221% year-over-year and a 54% quarter-over-quarter growth. In terms of specific delivery progress, Broadcom delivered TPU v7 to both Anthropic and Google, and TPU v8i to Google. Additionally, the company officially began delivering its custom chip “Jalapeño” for OpenAI.Looking ahead to the upcoming fourth quarter, Broadcom expects further acceleration, including faster delivery of TPU v7 to Anthropic, increased production speed for Google’s TPU v8i, and the start of deliveries of MTIA series accelerators to Meta. Based on this, Broadcom forecasts AI revenue in the fourth quarter to rise to $21.7 billion, with an estimated year-over-year increase of 236%. The total AI revenue for the entire fiscal year 2026 is expected to reach $58 billion, a significant increase of 186% compared to the same period last year.From a specific capacity deployment plan perspective, the investment pace of major players is accelerating across the board: Broadcom and Anthropic plan to deploy 1 GW of TPU v7 in 2026, expand to 5 GW of TPU v8i in 2027, and add another 10 GW in 2028. Meanwhile, OpenAI plans to deploy 1.3 GW of the “Jalapeño” chip in 2027, expanding the deployment to over 5 GW by 2028 (including the “Jalapeño” and its subsequent upgrades). Both parties are also steadily advancing the development of the third-generation XPU. In addition, Broadcom plans to deliver three generations of MTIA accelerators to Meta by the end of 2027, with the total deployment scale expected to reach 3 GW by the end of 2028.With strong order reserves and clear capacity planning, Broadcom is confident about future market performance. The company expects its AI semiconductor business revenue to jump to approximately $115 billion in 2027 and $230 billion in 2028, achieving continuous doubling growth, with the upstream supply chain fully prepared.

Join NowENLatest AI NewsHuang Renxun urges at G20: Don’t fear and regulate AI too much, being left behind by the times is the worst outcomeNVIDIA CEO Huang Renxun called on countries to embrace AI infrastructure and applications at the G20 Ministerial Meeting on Innovation, warning that excessive fear or regulation could miss the opportunity of the next round of technological revolution. He suggested treating AI as an essential economic foundation, comparable to electricity, roads, and the internet, and promoting its construction like water and power.just now0AI Assisted 3 Weeks to Finish 12 Years of Work: Paint.NET First Lands on Linux Using Self-Developed Solution to Bypass Direct2DDeveloper Rick Brewster released Paint.NET 5.2 Build 9739, adding first experimental Linux support via Wine. The main obstacle was Windows-only Direct2D API, which Paint.NET long relied on. Since Wine’s Direct2D support was incomplete, he built an alternative, ending a 12-year blocker…..just now4.7KBaidu’s Basic Model Department Undergoes Major Changes: North American Large Model Experts and Former DeepSeek Core Members Join, Wenshi Multimodal Technology Sees Key UpgradesBaidu’s foundation model unit BMU is ramping up talent: hiring Frontier Lab researcher Wu Qin to boost Ernie pre-training and naming ex-DeepSeek researcher Wei Haoran head of Ernie multimodal algorithms, as new chief Sun Tianxiang drives Ernie upgrades…..just now5.9KTmall Officially Launches AI Space Station: Alibaba Cloud, Zhipu, Kimi, and MiniMax Jointly JoinOn Sept 3, Tmall launched AI Space Station (Token Center), first integrating top LLM subscriptions from Alibaba Cloud, Zhipu, Kimi, and MiniMax into e-commerce, letting users buy AI as easily as phone top-ups, marking a new milestone for consumer LLM adoption in China…..just now7.6KBroadcom predicts Anthropic and OpenAI will surpass Google to become the top two customers for AI custom chipsPatrick Chi, president of Broadcom, predicts that as large model vendors intensify their competition for computing power, Anthropic and OpenAI will surpass Google to become the top two core customers for Broadcom’s AI ASIC business. Anthropic is expected to become the largest XPU custom chip customer by 2027 to 2028, followed closely by OpenAI in second place.just now5.4KAI Performance Evaluation Shifts from Usage to Output: Meta Abandons Adoption Dashboards and Token MetricsA memo from Meta executives clearly states that AI adoption dashboards or token usage will not be used to evaluate engineers’ influence. Currently, 93% of code changes are assisted by AI agents, and AI is deeply embedded in engineering processes, but the emphasis is that using AI itself is not the ultimate goal. This move responds to the common practice in the tech industry of setting performance metrics to promote AI.just now7.0KAnthropic Releases Claude Fable5.1: Enhanced Agent Coding Capabilities, Up to 45% Cost ReductionAnthropic launched Claude Fable5.1 and restricted Mythos5.1, improving agentic coding and scientific reasoning while optimizing inference costs. Fable5.1 API is fully available on public cloud; Mythos5.1 is limited to cybersecurity and life sciences. Both are first Claude models with built-in AI text watermarking; Fable5.1 performs strongly in agentic tasks…..just now10.1KBaidu Brings in North American Large Model Expert, Former DeepSeek Researcher Now Head of Wenxin MultimodalBaidu’s Basic Model R&D Department (BMU) has completed key talent deployment: On September 3rd, the head Sun Tianxiang announced that senior researcher Wu Qin was recruited from a North American Frontier Lab to enhance the pre-training capabilities of the Wenxin Large Model; former DeepSeek researcher Wei Haoran has officially transferred to BMU as the head of Wenxin Multimodal Algorithm. His team previously open-sourced the Unlimited OCR model.just now10.0KAI Computing Power Boom Gives Rise to Super Large Orders: Nscale Secures a $45 Billion Order from Anthropic, Contract Revenue Surpasses the $100 Billion MarkAccording to The Information, after UK cloud provider Nscale secured a $45 billion compute deal with AI firm Anthropic, its disclosed total contract revenue rose from about $51 billion to about $103 billion, surpassing $100 billion for the first time…..just now9.7KMinistry of Justice Makes First Public Statement on AI Copyright: Supports OpenAI’s View that AI Training Constitutes Fair UseThe U.S. Justice Department filed a statement of interest in Manhattan federal court, backing OpenAI in NYT v. OpenAI. It argues training LLMs on copyrighted material is fair use, citing national security and transformative use. Signed by Deputy AG Woodward and others…..just now9.8KOfficial Launch of the Shuiyue Yuyue Dark Side Flat Magnetic Flagship Headphones, Price: 19999 YuanMoondrop’s flagship planar magnetic headphones “Dark Side of the Moon” launch Sept 2 at 19,999 yuan, targeting Hi-End. Features 100mm planar magnetic driver and 500nm ultra-thin d

Join NowENPremium Membership ·Limited-Time Pricing— Save TodayChoose a monthly plan that fits your needs. Each tier includes tailored points allowances and flexible usage limits for confident monitoring.Your current plan isFreeEssentialStart with essential GEO monitoring and analysis$14.9/30 daysGet PremiumPoints IncludedOne-time points 7,000Daily bonus points 100AI Query Mining50 usesGEO Rank Check50 uses5 AI platforms:GEO Brand Score Audit3 uses5 AI platforms:GEO Link Citation Tracking5 uses5 AI platforms:GEO Rank TrackingNot includedData Export：Not includedReport Generation：Not included7-Day Featured Slot in AI DirectoryNot includedDedicated SupportNot includedMost PopularGrowthBuilt for growing brands and daily GEO operations$45/30 daysGet PremiumPoints IncludedOne-time points 25,000Daily bonus points 200AI Query Mining150 usesGEO Rank Check100 uses7 AI platforms:GEO Brand Score Audit8 uses7 AI platforms:GEO Link Citation Tracking15 uses7 AI platforms:GEO Rank TrackingNot includedData Export：Not includedReport Generation：Not included7-Day Featured Slot in AI DirectoryNot includedDedicated SupportNot includedProProfessional GEO monitoring and continuous growth analysis$109/30 daysGet PremiumPoints IncludedOne-time points 65,000Daily bonus points 300AI Query Mining300 usesGEO Rank Check300 uses10 AI platforms:GEO Brand Score Audit20 uses10 AI platforms:GEO Link Citation Tracking30 uses10 AI platforms:GEO Rank TrackingUnlimited10 AI platforms:Data Export：IncludedReport Generation：Included7-Day Featured Slot in AI DirectoryNot includedDedicated SupportNot includedEnterprise PlanEnterpriseScalable GEO operations for enterprises and agencies$199/30 daysGet PremiumPoints IncludedOne-time points 120,000Daily bonus points 500AI Query Mining800 usesGEO Rank Check600 uses10 AI platforms:GEO Brand Score Audit30 uses10 AI platforms:GEO Link Citation Tracking60 uses10 AI platforms:GEO Rank TrackingUnlimited10 AI platforms:Data Export：IncludedReport Generation：Included7-Day Featured Slot in AI Directory1 usesDedicated SupportIncludedCompare PlansCompare usage allowances and premium benefits across plans at a glanceFree$0Get PremiumEssential$99/30 daysGet PremiumGrowth$299/30 daysGet PremiumPro$699/30 daysGet PremiumEnterprise$1,299/30 daysGet PremiumPoints IncludedOne-time points07,00025,00065,000120,000Daily bonus points100100200300500AI Query MiningQuery allowance550150300800GEO Rank CheckQuery allowance350100300600AI platformsGEO Brand Score AuditQuery allowance1382030AI platformsGEO Link Citation TrackingQuery allowance15153060AI platformsGEO Rank TrackingGEO Rank TrackingNot includedNot includedNot includedUnlimitedUnlimitedData ExportNot includedNot includedNot includedIncludedIncludedReport GenerationNot includedNot includedNot includedIncludedIncluded7-Day Featured Slot in AI Directory7-Day Featured Slot in AI DirectoryNot includedNot includedNot includedNot included1Dedicated SupportDedicated SupportNot includedNot includedNot includedNot includedIncludedFrequently Asked QuestionsCommon questions about Aibase memberships, credits, GEO monitoring, and billingWhat is the difference between monthly membership credits and daily login bonus credits?Monthly membership credits are the core AI computing allowance included with your membership and are issued once per membership benefit cycle.Daily login bonus credits are an additional benefit. They are automatically issued when a member first signs in to or visits Aibase on a given day, with no manual claim required.Daily bonus credits are valid only for the day they are issued and do not roll over to the next day.Do I need to manually claim daily bonus credits? Are missed days credited later?No manual claim is required. Daily bonus credits are automatically issued when you first sign in to or visit Aibase that day.If you do not sign in to or visit Aibase on a given day, that day’s bonus credits will not be issued and will not be credited retroactively.Daily bonus credits are an additional membership benefit and do not affect your regular monthly membership credits.Do monthly membership credits roll over to the next month?No. Membership credits are issued in 30-day benefit cycles.Unused membership credits expire at the end of the current benefit cycle and do not roll over into the next cycle.At the beginning of a new benefit cycle, the credit allowance for your current membership plan is issued again.If I buy a 3-, 6-, or 9-month membership, will all credits and usage quotas be issued at once?No. A multi-month purchase provides membership access for the selected duration; it does not issue all future credits and usage quotas upfront.Membership benefits are still managed in 30-day cycles. Credits are issued and tool usage quotas are reset at the start of each cycle.For example, a 3-month membership consists of three consecutive 30-day benefit cycles.In what order are different credit balances used?Credits are deducted in the following order: daily login bonus credits →

IT之家9 月 2 日消息，科技媒体 9to5Mac 昨日（9 月 1 日）发布博文，报道称 Meta 公司推出 Muse Voice Transcribe，这是其首个实时音频感知模型。开发者可通过 Meta Model API 调用该能力，定价为每 1000 分钟音频 3 美元（现汇率约合 20.2 元人民币），等同每小时 0.18 美元（现汇率约合 1.2 元人民币）。该模型在同一流程中整合流式自动语音识别、说话人分离与端点检测，用户说话时，系统可同步输出文字，无需等待完整录音结束后再单独处理。IT之家注：“流式”就是边说边转写，模型不必等整段音频说完才出结果，而是会一小段一小段持续输出文字，所以延迟更低，适合实时听写、会议记录、通话字幕。Muse Voice Transcribe 可在包含 20 余名说话者的录音中分离不同发言者，还能识别说话是否结束，从而自动确定一段输入的边界。模型训练覆盖 70 余种语言，其中 25 种语言在发布时完成验证。该模型支持长度超过 1 小时的音频，也支持句内或句间的自然语言切换。开发者还可通过语言、关键词和上下文偏置，提高特定语言、术语或场景下的识别效果。在技术方面，Meta 为兼顾实时响应和识别准确度，引入名为自适应延迟（adaptive delay）的动态决策机制。系统不会为所有词语采用固定的速度与准确度取舍，而是针对每个词判断需要额外接收多少音频后再输出识别结果。对于较容易识别的语音，系统可更快提交转写结果。对于发音不清、术语较多或语境复杂的词语，系统会调用更多前后文音频信息。这种机制旨在兼顾。Meta 称，截至 2026 年 9 月 1 日，Muse Voice Transcribe 位列 Artificial Analysis 流式语音转文本排行榜第 1 名。