AI NEWSLatest AI NewsArticleGoogle DeepMind releases WeatherNext 3, a meteorological large model that provides hourly updates and 5-kilometer high-precision global forecastsPublished in Latest AI NewsTime :Sep 4, 2026Read :6minuteRecently, Google DeepMind and Google Research jointly released the new global weather AI model WeatherNext3. This model breaks through the inherent bottlenecks of traditional meteorological AI in local terrain resolution and initialization delay by directly accessing real-time geostationary satellite mosaics for hourly initialization and using raw meteorological station observation data for training, achieving a comprehensive upgrade from data architecture to forecast accuracy.In terms of core architecture and input mechanism, WeatherNext3 adopts the same probabilistic functional generation network (FGN) grid Transformer architecture as WeatherNext2 and further expands to multi-resolution output. Its input directly connects to global real-time geostationary satellite mosaics and the HRES analysis data from the European Centre for Medium-Range Weather Forecasts (ECMWF). During the training phase, it extensively absorbs multi-source information including ERA5/HRES-fc0, NASA’s IMERG, meteorological station measured data, and satellite mosaics. Traditional meteorological AI models usually rely on reanalysis grids from numerical weather prediction (NWP), which smooths out local climate changes caused by coastlines, valleys, and mountains. WeatherNext3, however, trains a dedicated output head using raw meteorological station observation data, making its 0.05-degree temperature and dew point forecasts more consistent with actual instrument records rather than relying solely on the model’s own physical atmospheric representation.In terms of resolution and forecast frequency (Cadence), WeatherNext3 can output three levels of forecast results with a single forward pass. The first level is 2-meter air temperature and dew point at a resolution of 0.05 degrees (approximately 5 kilometers) based on meteorological station training. The second level is grid surface data at a resolution of 0.1 degrees (approximately 10 kilometers), including 10-meter and 100-meter wind speed, pressure, sea surface temperature, cloud distribution, solar radiation, and one-hour precipitation. The third level is an atmospheric field covering 13 pressure levels at a resolution of 0.25 degrees (approximately 25 kilometers). Compared to the previous model’s output of 0.25-degree fields every six hours, its accuracy has improved by about five times. In terms of update frequency, the model is initialized 24 times a day. The weather system cycles at UTC 00, 06, 12, and 18 can extend the forecast up to 15 days (360 hours) and include 64 ensemble members, while temporary hourly runs cover a 48-hour window. This hourly rolling refresh based on current satellite observations completely breaks the previous cycle limitations that relied on lagging six-hour traditional numerical forecast analysis, offering significant value for capturing rapidly developing convective weather.In terms of precipitation simulation and support for clean energy variables, WeatherNext3 demonstrates strong business application potential. Precipitation forecasting has always been a historical pain point for global models, often producing unclear fields that miss storm boundaries. This model trains using multi-source data including ECMWF reanalysis, NASA’s IMERG satellite inversion, and Google’s own satellite radar precipitation reanalysis. In the early forecast stage, its Continuous Ranked Probability Score (CRPS) can improve by up to 60% compared to the IMERG benchmark and shows significant improvement compared to meteorological station rain gauges. For the new energy sector, the model can also accurately output 100-meter wind speed at the hub height of wind turbines, complete distributions of low, medium, and high clouds, and all components of solar irradiance. These key data combinations can directly empower power grid operators to accurately predict wind and solar energy generation output.Currently, the forecast data of WeatherNext3 is open for application, and users can access it through BigQuery, Earth Engine, and Cloud Storage. However, the model weights are not open-sourced, and custom inference is still running the previous version of WeatherNext2. As this model is fully implemented, global weather forecasting is accelerating towards a new intelligent era of high precision and high timeliness.

AI NEWSLatest AI NewsArticleAI-Generated Menus Fall into Perfectionist Aesthetics, Food Photos Trigger the Uncanny Valley EffectPublished in Latest AI NewsTime :Sep 4, 2026Read :3minuteGenerative AI is entering the design of restaurant menus, but its highly homogenized “perfect aesthetics” has also raised new issues with visual experience. More and more restaurants are using AI-generated images for food menus, which often exhibit overly symmetrical, smooth, and refined characteristics. Some works even show obvious problems such as abnormal food structures, which consumers describe as having a “uncanny valley effect.“Alex Lisle, Chief Technology Officer of Reality Defender, believes this phenomenon is related to the training mechanism of AI models. LLMs and diffusion models need to learn patterns from massive data, and when models frequently refer to similar-style chain restaurant menus, the generated results tend to become similar; if AI-generated content further enters the training data, it may reinforce this style convergence. Lee Renney, Director of the “Center for Imagining Digital Futures” at Elon University, stated that training data usually tends to be aesthetically pleasing and avoids offensive content, which further pushes images and language towards being “rounded off.“This issue may become more apparent with repeated editing. A user found that after continuously modifying an AI-generated menu 100 times in ChatGPT, the food images gradually deviated from their original form. Researchers from the University of Duisburg-Essen previously found that AI-generated food images may produce a negative perception similar to the “uncanny valley.“Industry insiders point out that AI-generated content is entering the real information environment on a large scale, and the data generated by the model itself may in turn affect subsequent training. How generative AI can avoid aesthetic homogenization and maintain the details and diversity of the real world will become an important issue beyond the model’s capabilities.

AI NEWSLatest AI NewsArticleSmart Ring Manufacturer Oura Has Officially Filed for IPO, 9-Month Revenue Increased to $1.2 BillionPublished in Latest AI NewsTime :Sep 4, 2026Read :4minuteSmart ring manufacturerOura has officially submitted its IPO application. The company filed documents with the U.S. Securities and Exchange Commission (SEC) on Thursday, showing that Oura’s revenue increased from$697 million to $1.2 billionover the past nine months ending June 30. The company previously projected revenue of about $500 million in 2024, $1 billion in 2025, and nearly $2 billion in 2026.As of now, Oura has sold approximately3.6 million smart ringsin the past year, with around5 million paid members, and a 12-month weighted average member retention rate of about 85%. Oura rings are priced between $350 and $400, covering physiological indicators such as heart rate, metabolism, stress, and sleep, and provide continuous health monitoring services through its companion app.Oura was founded in Finland in 2013 and submitted a confidential IPO application in May this year. Earlier market reports indicated that the company planned to raise about3 billion dollarsthrough the IPO, with a potential valuation reaching16 billion dollars; the company’s valuation was approximately 11 billion dollars in October 2025.Oura stated that it will expand from traditional activity and fitness tracking to broader healthcare scenarios, and deepen cooperation with insurance companies, employers, and healthcare providers. The company disclosed that it has accumulated more than50 health indicators and nearly 4.2 billion hours of physiological data, and will use these longitudinal data to train AI and machine learning models to enhance health pattern recognition, personalization, and predictive capabilities.Meanwhile, Oura is facing a proposed class-action lawsuit alleging that it misled consumers regarding the accuracy of its sleep tracking. Oura denies the allegations and states it will defend itself through legal means. This IPO will serve as an important window for observing the evolution of smart wearables from hardware sales to a “hardware + subscription + AI health service” model.

AI NEWSLatest AI NewsArticleOverseas Top Three AI Systems Experienced Near 4-Hour Downtime, Zhipu Announced That They Are Still Online and Have Launched a Nighttime Free ServicePublished in Latest AI NewsTime :Sep 4, 2026Read :4minuteLeading AI services in the United States experienced a collective failure, with ChatGPT, Claude, and Grok announcing service abnormalities, while Google’s Gemini and Microsoft Copilot also received numerous interruption reports. This large-scale outage lasted approximately 3 hours and 40 minutes, with the failure first erupting around 9:23 AM Eastern Time on September 3rd.Claude was the first to report errors, with OpenAI receiving over 37,000 fault reports at peakAccording to data from the overseas fault monitoring platform Down Detector, the Claude series models from Anthropic experienced a sharp rise in error rates, including core models such as Mythos 5.1, Opus, and Fable 5.1. Just two minutes later, xAI’s Grok service went offline entirely; about an hour and a half later, OpenAI’s ChatGPT and programming tool Codex also experienced large-scale access errors.During the peak of the incident, more than 37,000 fault reports were submitted globally for OpenAI services, with 80% concentrated on ChatGPT. At the same time, Google’s Gemini and Microsoft Copilot received numerous user interruption feedback, while the AI programming tool Cursor directly announced that some of its services had been forced to interrupt due to upstream large model failures. By 11:16 AM Eastern Time on the same day, Claude’s main services recovered first, followed by gradual restoration of ChatGPT and Grok.Domestic large models remained stable, and Zhipu launched free nighttime usageNotably, during the period when overseas large models collectively failed, domestic large models remained stable. That evening, Zhipu posted on social media X: “We are still up."—which translates to “We are still online.“At the same time, Zhipu announced on the evening of September 3rd that from now until September 20th, GLM-5.3-Flash in Z Code would be available for free between 11 PM and 9 AM daily, with the usage quota doubled for other Agents supported by the package, automatically taking effect during the specified time period.According to the introduction, GLM-5.3-Flash is the first native multimodal model in the GLM-5 series, scoring 57 points in the comprehensive intelligence index of the globally authoritative Artificial Analysis, entering the global cutting-edge model capability range. This model is suitable for code generation, code understanding, problem fixing, and other code tasks, and shows significant improvements in Office and document-related tasks, as well as specialized tasks in finance and law.

Join NowENLatest AI NewsMusk’s daughter shoots an ad, treating a robot as a dog, and even confronts AI taking away human jobsVivian Wilson, 22-year-old daughter of Musk, becomes the autumn 2026 ambassador for Spanish brand Desigual, with the theme “naturally rebellious.” In the ad, she playfully challenges AI, giving absurd instructions to the DESI 84, a humanoid robot inspired by Tesla: use it as a golf cart, walk it like a pet, command it to mow the lawn, and even have an interactive scene on the bed.1 days ago12.8KIsrael-Netherlands AI Unicorn Wonderful Secures $550 Million C Round Financing, Valuation Surges to $5 BillionIsraeli-Dutch AI startup Wonderful closed a $550M Series C at a $5B valuation, more than doubling from about $2B under six months ago, with leading investors increasing stakes…..1 days ago14.8KOverseas Top Three AI Systems Experienced Near 4-Hour Downtime, Zhipu Announced That They Are Still Online and Have Launched a Nighttime Free ServiceTop US AI services were down about 3h40m. ChatGPT, Claude, and Grok had errors; Google Gemini and Microsoft Copilot saw many outage reports. The incident began around 9:23 a.m. ET on Sept. 3, with Claude models failing first. At peak OpenAI had over 37,000 outage reports…..1 days ago18.6KSmart Ring Manufacturer Oura Has Officially Filed for IPO, 9-Month Revenue Increased to $1.2 BillionOura filed for IPO. Revenue in the 9 months ended June 30 reached $1.2B, up from $697M a year earlier. It projects 2024 revenue ~$500M, 2025 ~$1B, 2026 nearly $2B. Sold ~3.6M rings in the past year, ~5M paying members, ~85% retention, priced $350-$400…..1 days ago15.7KAI-Generated Menus Fall into Perfectionist Aesthetics, Food Photos Trigger the Uncanny Valley EffectGenerative AI in menu design brings homogenized “perfect aesthetics” causing visual issues. AI dish images are overly symmetrical and polished, with unnatural food structures, called “uncanny valley effect.” Experts attribute this to AI models learning patterns from massive data and averaging outputs…..1 days ago15.5KGoogle DeepMind releases WeatherNext 3, a meteorological large model that provides hourly updates and 5-kilometer high-precision global forecastsGoogle DeepMind and Google Research jointly released global weather AI model WeatherNext3, overcoming bottlenecks in local terrain resolution and initialization latency by directly ingesting real-time geostationary satellite mosaics for hourly initialization and training on raw station observations, comprehensively improving forecast accuracy. Its core architecture continues using the probabilistic functional generative network (FGN) grid T from ….1 days ago16.0KNVIDIA’s $1.29 Billion Acquisition of Hugging Face Huang Renxun Makes New CommitmentNvidia announced its acquisition of Hugging Face, an open-source AI platform, for $12.9303 billion (~87.092 billion yuan), drawing tech attention. CEO Jensen Huang explained the rationale and plans. Hugging Face has gathered over 18 million developers and researchers…..1 days ago15.4KGPT Image 2.5 Multiple Upgrades Attract Attention, Realistic Images Spark New ExpectationsThe unreleased GPT Image 2.5 has drawn tech community attention, with version jumping from speculated 2.1 to 2.5, signaling fast iteration and technical ambition. Suspected to be LMArena’s anonymous model luna-lisa-alpha, it shows major improvements in image generation, faster speed and better visual quality…..1 days ago16.4KGoogle DeepMind Releases WeatherNext 3: 5-Kilometer Resolution Global Forecast Updated Every HourGoogle DeepMind and Google Research release WeatherNext 3 to address coarse resolution, poor local terrain representation, and ~6h NWP initialization lag. It trains directly on station observations, uses real-time global geostationary satellite mosaics, and reinitializes hourly, improving forecast resolution and timeliness…..1 days ago15.6KGoogle Assistant to be Discontinued Starting September 4, Google Gemini to Take OverGoogle confirmed Google Assistant on Android will gradually stop from Sept 4, migrating users to Gemini. Email notices sent; migration may take weeks, some delayed, but all will lose access. Originally planned for late 2025, delayed to refine Gemini, now deemed ready. Smartwatches etc. also affected…..1 days ago13.4KAI Data Center Star Crusoe Raises $3 Billion: Valuation $30 Billion, Just Signed a $13 Billion DealData center developer Crusoe has completed a new $3 billion funding round, bringing its valuation to $30 billion. Its clients include Meta, Microsoft, and OpenAI. The round was led by Atreides Management and Valor Equity Partners, with Mubadala Capital participating. Just 10 months since its last funding round, the valuation has tripled from $10 billion.1 days ago14.0KGlobal AI Strike: Overseas Main Large Models Suffered a System Failure for Over 3 Hours, Domestic Large Models Remain Online CalmlyOn Sept. 3 around 9:23 a.m. ET, major overseas AI services including ChatGPT, C

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 →

Claude Fable 5.1 – Anthropic 推出的最新旗舰大模型AI工具3天前更新AI小集02Claude Fable 5.1是什么Claude Fable 5.1 是 Anthropic 推出的旗舰大模型，定位为目前公开可用的最强 AI 之一。模型专为复杂推理、长周期 Agent 任务和知识工作设计，在科学研究、编码、计算机操作等 8 项基准测试中全面领先上一代，且低推理档位即可达到前代高档位性能。同时缓存读取成本降低 75%，典型工作负载成本下降约 25%，高度 Agent 化任务最高省 45%，实现更强性能与更高性价比的统一。Claude Fable 5.1的主要功能复杂推理：专为多步骤深度推理设计，能追踪问题根本原因并稳定执行数小时甚至数十小时的长周期任务。Agent工作流：支持自主规划、工具调用与多步骤任务编排，适用于编码、科研发现及业务流程自动化。科学研究：在蛋白质设计、天文学数据分析（如基于NASA雷达图像生成金星高分辨率地形图）和计算生物学等领域具备实战能力。软件开发：在多项编码基准测试中领先，可定位罕见系统崩溃根源、优化代码性能并支持长时间开发会话。计算机操作：具备GUI理解与自动化操控能力，可在OSWorld等环境中完成复杂的跨应用任务。知识工作：处理文档分析、写作、多学科推理等知识密集型任务，对风格提示词遵循度更高。Claude Fable 5.1的技术原理共享基础架构：与Mythos 5.1基于同一底层模型，通过差异化安全措施实现分级访问，Fable面向公众开放，Mythos仅向审核通过的网络安全与生命科学机构提供更高能力边界。可调推理强度：内置low至max多档推理强度，用户可根据任务复杂度在响应速度、API成本与准确率之间动态权衡，低中档位即可达到前代高档位性能。上下文缓存优化：通过将缓存读取成本降低75%，使模型在复用已处理上下文时显著削减高度Agent化任务的整体开销，典型负载成本降约25%，重度Agent任务最高降45%。反蒸馏保护：为每个思维链区块附加签名，验证对话前缀未被篡改；若上下文被编辑则校验失败，直接阻断通过篡改历史对话提取模型推理能力的蒸馏攻击路径。长程稳定性机制：改进多步骤任务中的错误传播控制，避免早期微小错误在后续数十步中放大导致全盘崩溃，同时能在遇阻时清晰汇报已尝试方案与卡点位置。企业隐私架构（EFS）：将数据存储于客户自有云基础设施而非Anthropic服务器，由客户而非Anthropic执行必要的人工审核，在维持安全监控的同时实现零数据保留级别的隐私保护。如何使用Claude Fable 5.1API 调用：开发者通过 Anthropic API 指定模型 IDclaude-fable-5-1可集成到自有应用或工作流中。网页端聊天：直接访问Claude，在对话界面选择 Fable 5.1 进行日常问答、写作与知识工作。Claude Code：在命令行工具中使用，默认开启高推理强度，适合复杂编码与长周期开发任务。Claude Cowork：在协作场景中调用，默认使用中等推理强度，平衡响应速度与任务质量。云平台部署：通过 Amazon Bedrock、Google Cloud 或 Microsoft Azure 等云服务商接入，满足企业级基础设施需求。调节推理强度：根据任务难度在 low、med、high、xhigh、max 五档间切换，以控制成本与响应延迟。Claude Fable 5.1的核心优势性能全面领先：在科学研究、编码、知识工作等 8 项公开基准测试中全面超越前代 Fable 5 与 GPT-5.6 Sol，树立新标杆。极致性价比：缓存读取成本降低 75%，典型工作负载总成本降约 25%，高度 Agent 化任务最高可省 45%。低档位高性能：低、中推理档位即可达到前代高甚至极高档位的表现，用更少资源完成更难任务。长周期稳定性：擅长执行持续数小时甚至数十小时的复杂多步骤任务，能追踪问题根本原因并防止早期小错误导致全盘崩溃。科研实战验证：在蛋白质设计（命中率近 50%）、金星高分辨率地形图生成、计算生物学 GPU 内核优化等领域取得可验证突破。分级安全架构：与 Mythos 5.1 共享底层能力但采用差异化安全措施，既向公众开放最强模型，又通过受控访问支持高风险专业研究。企业隐私保护：通过 Enterprise Frontier Safeguards (EFS) 将数据存储于客户自有云端，实现零数据保留级别的隐私安全。Claude Fable 5.1的项目地址项目官网：https://www.anthropic.com/claude-fable-and-mythos-5-1Claude Fable 5.1的同类竞品对比维度Claude Fable 5.1GPT-5.6 Sol科学研究Terminal-Bench-Science 0.1 得分52.6%22.4%Agent 编码Terminal-Bench 4.0 得分55.8%37.3%知识工作GDPval-AA v2 得分18531711业务流程AutomationBench 得分31.4%19.6%计算机操作OSWorld 2.0 有数据（77.9% partial）无公开数据性价比缓存读取降 75%，典型负载省 25%无同等降价Claude Fable 5.1的应用场景复杂软件开发：能追踪并定位困扰工程团队数年的罕见系统崩溃根源，支持长达数十小时的连续编码与多实验并行任务。生命科学蛋白质设计：调用开源工具设计高亲和力蛋白质结合剂，在12个靶标上实现近50%的实验验证命中率，亲和力达行业竞赛最佳方案的10倍。天文学数据分析：基于NASA麦哲伦号30多年前的雷达图像，为金星约三分之一表面生成分辨率提升至2-3公里的高程图，精度提高25%。企业业务流程自动化：执行跨系统的复杂多步骤业务工作流，在AutomationBench基准上得分较上一代提升近一倍，从17.1%跃升至31.4%。计算基础设施优化：自动编写自定义GPU内核并缓存中间结果，将7个开源深度学习模型推理速度最高提升2.5倍，降低30%-60%的GPU计算成本。# AI工具# AI项目和框架©版权声明本站文章版权归AI工具集所有，未经允许禁止任何形式的转载。上一篇WeMM-Embedding - 腾讯开源通用多模态Embedding模型下一篇Claude Mythos 5.1 - Anthropic 推出的 Claude 系列旗舰模型相关文章Baichuan4-Finance – 百川智能推出的全链路金融领域增强大模型AI小集3Command A+ – Cohere Labs 开源的多模态大模型AI小集2Seed1.6 – 字节跳动推出的通用模型系列AI小集3悦颜相机 – AI拍照和图像编辑应用，制作动态写真效果AI小集2Animated Drawings – Meta AI推出的AI手绘作品转动画工具AI小集5Muse Spark 1.3 – Meta 推出的新一代大语言模型AI小集2暂无评论再想想发表评论暂无评论…热门工具Loomy即梦AISekoAiPPT秘塔AI搜索妙呀Lovart美图设计室绘蛙AIupdream办公小浣熊秒哒最新收录Tokera AI文优小助ByteCPAgent.SpaceQwenWorkDeepSkill最新文章秒悟Meoo – 1分钟生成前后端完整的网站，真0门槛开发！2小时前Orbis – Visko推出的实时世界模型，让AI视频像直播一样2小时前SchoAI思果学术 – 科研学术 AI 助手，一键完成论文写作19小时前Pi – 开源的终端编程 Agent，支持自定义工具与模型接入23小时前E-Commerce Bench – 千问联合淘天开源电商经营评测基准1天前zg – Zvec AI 开源的本地优先统一检索基础工具1天前MAI-Transcribe-2 – 微软推出的最强 AI 语音转文字模型1天前GPT-6 Astra – OpenAI 推出的最新旗舰大模型1天前Cheso – 腾讯推出的 AI PPT Agent 制作工具2天前Muse Spark 1.3 – Meta 推出的新一代大语言模型2天前Easel – 浙大联合北大开源的 AI 社交媒体内容工作台2天前Gemini 3.8 – 谷歌推出的推理与编码模型2天前Fusion Model – FUMO Lab 推出的智能分配模型3天前Hy4 preview 轻量版 – 腾讯混元开源的Hy4量化压缩模型3天前Muse Voice Transcribe – Meta 推出的首个实时音频感知模型3天前

Muse Spark 1.3 – Meta 推出的新一代大语言模型AI工具2天前更新AI小集02Muse Spark 1.3是什么Muse Spark 1.3 是 Meta 推出的新一代大语言模型，由扎克伯格亲自站台，定位为 Meta 在编程和智能体任务上的重大突破。模型在代码理解与长文本处理方面表现尤为出色，工具调用和 Token 消耗显著降低，同时用极具竞争力的价格向开发者开放 API。Muse Spark 1.3的主要功能长任务 Agent 工作流：支持长时间运行的开放式任务，能自主收集上下文、从杂乱信息中整理线索、发现计划缺口后主动调整，持续记录状态直至交付。编程能力增强：针对长时间编程任务优化，减少不必要的对话轮次，代码风格更干净，在代码库理解与终端编程上表现突出。复杂指令遵循：在多步骤、多限制条件的任务中能持续记住前文要求，减少后半段遗漏条件或偏离流程的问题。长上下文处理：支持 1M tokens 上下文窗口，长文本检索准确率极高。人机协作优化：面对模糊指令主动追问用户，任务受阻时主动求助，涉及不可逆操作前会确认执行。成本效率提升：完成相同任务时工具调用减少约 20%，Token 消耗减少约 25%。Muse Spark 1.3的技术原理后训练与测试时扩展：基于 Avocado 模型进行再次后训练，同时强化测试时扩展能力（test-time scaling），使模型在推理阶段能动态分配更多计算资源提升输出质量。多环境 Agent 训练：模型在多种不同的 Agent harness 中进行了训练，让它离开特定环境后依然能保持相近的工作能力。能力边界对齐：Meta 专门强化模型对自身能力边界的感知训练，使其能更准确地区分「知道什么、不知道什么、能做什么、做不了什么」，遇到障碍时明确表达而非虚构结果。长程编程任务数据：增加大量长时间编程任务的训练数据，针对日常工程场景中的可用性进行了优化，使模型在真实开发工作流中更高效、更克制。如何使用Muse Spark 1.3Muse Code终端编程：在终端执行curl -fsSL https://dev.meta.ai/install.sh | bash安装 Muse Code，可在编程工作流中直接调用 Muse Spark 1.3 作为底层模型。Meta Model API 接入：开发者通过官方 API 直接调用，标准定价为输入 $1.25/百万 tokens、输出 $4.25/百万 tokens。Contributor 折扣档：安装 Muse Code 后默认开启，以输入 $0.10、输出 $0.20 的低价使用，代价是提示词和代码输出会被 Meta 用于训练，可手动切换退出。第三方平台集成：Vercel AI Gateway 等合作伙伴平台已接入 Muse Spark 1.3，方便在现有工作流中快速集成调用。Muse Spark 1.3的核心优势编程能力领先：在 DeepSWE、SWEAtlas、Terminal-Bench 等编程基准中多项第一，超越 GPT-5.6 Sol 和 Claude Opus 5。长上下文极强：支持 1M tokens 上下文，长文本检索准确率高达 98% 以上，大幅领先竞品。成本效率双优：相同任务下工具调用减少约 20%、Token 消耗减少约 25%，API 定价远低于同级别模型。Agent 长任务自主化：能自主收集信息、发现计划缺口后主动调整，在复杂开放式任务中持续保持状态。人机协作更自然：面对模糊指令主动追问，任务受阻时主动求助，涉及不可逆操作前会确认执行。性价比极具竞争力：标准档定价仅为 Claude Opus 5 的约 1/5，Contributor 折扣档更是低至”几乎忽略不计”。Muse Spark 1.3的同类竞品对比对比维度Muse Spark 1.3Claude Opus 5发布方MetaAnthropic上下文窗口1M tokens200K tokens编程能力DeepSWE v1.1 得分 75.4%，领先DeepSWE v1.1 得分 74.0%，稍逊知识工作GDPVal-AA v2 得分 1754GDPVal-AA v2 得分 1824，更强计算机操控OSWorld 2.0 得分 66.9OSWorld 2.0 得分 68.3，略优标准输入定价$1.25 / 百万 tokens$5 / 百万 tokens标准输出定价$4.25 / 百万 tokens$25 / 百万 tokensMuse Spark 1.3的应用场景智能编程开发：通过 Muse Code 完成复杂代码编写、跨文件调试与大型代码库理解，工具调用和 Token 消耗显著降低。长文档深度分析：利用 1M tokens 上下文窗口处理百万字级长文本，进行跨章节信息提取、关联分析与精准检索。个人 Agent 长期任务管理：持续跟踪用户目标，自主收集信息、发现计划缺口后主动调整策略，直至交付最终成果。复杂工程报告生成：整合多源异构数据（如 CFD 仿真结果、CAD 模型），按指定格式自动生成结构化专业报告。# AI工具# AI项目和框架©版权声明本站文章版权归AI工具集所有，未经允许禁止任何形式的转载。上一篇Easel - 浙大联合北大开源的 AI 社交媒体内容工作台下一篇Cheso - 腾讯推出的 AI PPT Agent 制作工具相关文章Mercury – Inception Labs推出的扩散语言模型AI小集3Meku – AI开发平台，简单描述生成Web应用和网站AI小集3Logome – AI LOGO 设计工具，根据品牌信息提供智能设计建议AI小集2方舟 CLI – 字节跳动火山引擎推出的命令行工具AI小集2I2VGen-XL：阿里推出的图像到视频生成模型AI小集3podlm-public – 开源的AI博客生成工具，支持网页URL、文本转换成博客AI小集2暂无评论再想想发表评论暂无评论…热门工具Loomy即梦AISekoAiPPT秘塔AI搜索妙呀Lovart美图设计室绘蛙AIupdream办公小浣熊秒哒最新收录Tokera AI文优小助ByteCPAgent.SpaceQwenWorkDeepSkill最新文章秒悟Meoo – 1分钟生成前后端完整的网站，真0门槛开发！49秒前Orbis – Visko推出的实时世界模型，让AI视频像直播一样6分钟前SchoAI思果学术 – 科研学术 AI 助手，一键完成论文写作18小时前Pi – 开源的终端编程 Agent，支持自定义工具与模型接入21小时前E-Commerce Bench – 千问联合淘天开源电商经营评测基准1天前zg – Zvec AI 开源的本地优先统一检索基础工具1天前MAI-Transcribe-2 – 微软推出的最强 AI 语音转文字模型1天前GPT-6 Astra – OpenAI 推出的最新旗舰大模型1天前Cheso – 腾讯推出的 AI PPT Agent 制作工具2天前Easel – 浙大联合北大开源的 AI 社交媒体内容工作台2天前Gemini 3.8 – 谷歌推出的推理与编码模型2天前Fusion Model – FUMO Lab 推出的智能分配模型3天前Hy4 preview 轻量版 – 腾讯混元开源的Hy4量化压缩模型3天前Muse Voice Transcribe – Meta 推出的首个实时音频感知模型3天前Atlas – World Labs 推出的全球首个多模态世界模型3天前

感谢IT之家网友乌蝇哥的左手、不一样的体验的线索投递！IT之家9 月 3 日消息，Meta Platforms 于 2026 年 9 月 2 日发布迄今最强大 AI 模型 Muse Spark 1.3，专为延长智能体工作流与增强编码能力设计。Meta 首席 AI 官 Alexandr Wang 表示，该模型在编码能力上超越 OpenAI 的 GPT-5.6 Sol，与 Anthropic 的 Claude Fable 5.1 表现持平。定价方面，Muse Spark 1.3 维持与前代 Muse Spark 1.2 相同的定价，通过 Meta Model API 向开发者提供付费访问：每 100 万词元输入费用为 1.25 美元（IT之家注：现汇率约合 8.4 元人民币）每 100 万词元缓存命中输入费用为 0.15 美元（现汇率约合 1 元人民币）每 100 万词元输出费用为 4.25 美元（现汇率约合 28.6 元人民币）Wang 透露，API 平台市场采用强劲，部分开发者周用量达万亿级 token，新模型将逐步整合至 Instagram、Facebook 及 Meta AI 助手。官方博文指出相比较 Muse Spark 1.2，最新 1.3 减少了不必要的代码迭代，代码更简洁，整体编码风格也更加清晰，工具调用次数减少了约 20%，完成相同任务所需 token 减少 25%。Muse Spark 1.3 针对长周期智能体任务优化。模型能在单一长线程中协作处理多工作流，主动修正计划缺口并跟踪学习成果。面对模糊指令时，它会提出澄清问题；遇到障碍时请求用户帮助；执行关键操作前会寻求确认。新模型遵循复杂长指令的可靠性提升，多步骤任务中更好地保留详细需求，不丢失约束条件。多任务处理能力增强，能更准确地将新指令映射到单一线程中的正确任务。模型对自身能力与局限性的认知更强，减少幻觉输出。

Gemini 3.8 – 谷歌推出的推理与编码模型AI工具2天前更新AI小集02Gemini 3.8是什么Gemini 3.8 是 Google 推出的推理与编码模型，包含 Flash 和 Flash Cyber 两个版本。Gemini 3.8 Flash 在保持与前代相同速度和低价的基础上，大幅提升软件工程、Agent 任务及多步推理能力，在代码、金融、法律等基准测试中接近甚至超越更大参数的前沿模型。Gemini 3.8 Flash Cyber 是专为网络安全打造的版本，具备顶尖漏洞发现与自动补丁修复能力，仅通过 Fairwind Program 向认证安全防御人员开放。Gemini 3.8的主要功能长周期软件工程：自主完成复杂工程问题的端到端解决，在 DeepSWE 等基准中表现接近甚至超越更大参数的前沿模型。自主 Agent 任务：支持复杂多步骤的 Agent 工作流，可执行迭代式工具调用与推理，适用于终端编码、计算机操作等场景。专业领域推理：在金融分析、法律工作流及跨学科专家推理中展现出企业级可靠性。网络安全攻防：Flash Cyber 版本专注于自主漏洞发现与自动化补丁修复，在 CyberGym 和 CWE-Bench 等行业基准中达到前沿水平。原生多模态：支持文本、代码、图像、视频等多种模态输入，可生成 3D 可视化、游戏、交互式应用等富媒体内容。Gemini 3.8的技术原理长期运行的 Agent 循环机制：Gemini 3.8 引入长期运行的 Agent 循环（long-running agentic loops），该机制使模型能在任务执行过程中递归地评估和精炼自身输出。与单次前向传播不同，模型会在复杂任务中持续迭代——执行额外的推理步骤、反复调用外部工具、根据中间结果动态调整策略，在多步骤推理和代码生成任务中实现更深度的执行和更高的准确率。动态计算投入与推理强度调节：模型采用”投入更多计算量”的设计哲学，在复杂任务上展现出更高的执行勤勉度（greater diligence）。模型会根据任务难度自动分配更多推理资源，在更高推理强度（higher effort levels）下可能消耗更多 Token 以最大化性能；同时，开发者可选择较低的推理强度级别来减少 Token 开销，在性能与成本之间灵活权衡。网络安全领域的深度专项训练：Gemini 3.8 Flash 和 Flash Cyber 共享同一套底层基础智能，显著的代码与推理能力提升部分源于在网络安全这一高要求领域进行的严格训练。通过在漏洞发现、补丁修复等任务上的深度优化，模型学会了识别和修复跨 20 种编程语言的复杂代码漏洞，内化防御者优先的安全思维。如何使用Gemini 3.8开发者：通过Google AI Studio、Gemini API 或 Android Studio 调用 3.8 Flash，也可在Google Antigravity中探索 Agent 优先的工作流，或在 Stitch 中生成 UI。企业用户：直接在 Gemini Enterprise 平台中接入 3.8 Flash，用于构建企业级智能应用。普通消费者：订阅 Google AI Pro 或 Ultra 后，可在 Gemini App、Google Search 的 AI Mode 以及 Google Sheets 中使用 3.8 Flash。安全防御人员：通过 Fairwind Program 提交申请，获得认证后可优先访问 3.8 Flash Cyber，用于漏洞发现与自动化补丁修复。Gemini 3.8的核心优势极致性价比：用 Flash 系列的低价提供接近甚至超越 Claude Opus 5、GPT-5.6 等更大参数前沿模型的推理与代码能力。长周期软件工程：在 DeepSWE v1.1 等长周期软件工程基准中，能够自主端到端解决复杂工程问题，性能位居行业前列。企业级专业推理：在金融分析、法律工作流及跨学科专家推理等关键领域展现出支撑自主任务的可靠性。动态计算调节：支持根据任务复杂度灵活调整推理强度，既能在高难度任务上投入更多计算资源以最大化性能，也能在效率优先场景下降低 Token 消耗。网络安全专精：Flash Cyber 版本在漏洞发现和自动化补丁修复上达到前沿水平，且成本远低于同类大参数模型。原生多模态与 Agent 能力：基于长期运行的 Agent 循环机制，支持文本、代码、图像、视频等多模态输入，可自主迭代完成复杂多步骤任务。Gemini 3.8的项目地址项目官网：https://blog.google/innovation-and-ai/models-and-research/gemini-models/3-8-flash-and-3-8-flash-cyber/Gemini 3.8的同类竞品对比对比维度Gemini 3.8 FlashClaude Opus 5产品定位高智能”主力模型”，兼顾性能与成本旗舰级深度推理模型，追求极致智能输入价格$0.75 / 百万 Token$5.00 / 百万 Token输出价格$3.75 / 百万 Token$25.00 / 百万 TokenDeepSWE v1.1（长周期软件工程）73.7%74.0%HLE-Verified（跨学科专家推理）54.9%54.4%Harvey 法律 Agent（复杂法律工作流）10.0%6.7%Vals Finance Agent（金融分析）61.4%58.6%推理速度Flash 级别，高速响应相对较慢，侧重深度多模态能力原生支持文本、代码、图像、视频以文本推理为主核心优势极致性价比 + 自主 Agent 能力超长上下文 + 深度逻辑推理Gemini 3.8的应用场景自主软件开发：模型支持端到端完成复杂代码重构、Bug 修复与功能开发，支持长周期软件工程任务。金融量化分析：自动化解析财报与市场数据，生成投资研究报告、风险评估及合规建议。法律文档审查：快速分析合同条款、判例与法规，辅助律师完成尽职调查与合规审查。网络安全防御：自动扫描跨语言代码库漏洞并生成修复补丁，帮助安全团队快速响应威胁。多模态智能应用：基于自然语言生成 3D 交互游戏、数据可视化应用或富媒体内容。# AI工具# AI项目和框架©版权声明本站文章版权归AI工具集所有，未经允许禁止任何形式的转载。上一篇Fusion Model - FUMO Lab 推出的智能分配模型下一篇Easel - 浙大联合北大开源的 AI 社交媒体内容工作台相关文章VitaBench 2.0 – 美团 LongCat 推出的长期动态智能体评测基准AI小集2LongCat-Video-Avatar 1.5 – 美团开源的数字人视频生成模型AI小集2Kimi Claw – 月之暗面推出的云端AI智能体服务AI小集5Notehot.ai – AI写作和编辑平台，支持AI写作、内容优化、语法修复等多功能AI小集2Pic Answer – AI教育应用，拍照快速解答学术问题AI小集4GenChess – 谷歌DeepMind推出的定制个性化国际象棋棋子AI工具AI小集3暂无评论再想想发表评论暂无评论…热门工具Loomy即梦AISekoAiPPT秘塔AI搜索妙呀Lovart美图设计室绘蛙AIupdream办公小浣熊秒哒最新收录Tokera AI文优小助ByteCPAgent.SpaceQwenWorkDeepSkill最新文章秒悟Meoo – 1分钟生成前后端完整的网站，真0门槛开发！55秒前Orbis – Visko推出的实时世界模型，让AI视频像直播一样7分钟前SchoAI思果学术 – 科研学术 AI 助手，一键完成论文写作18小时前Pi – 开源的终端编程 Agent，支持自定义工具与模型接入21小时前E-Commerce Bench – 千问联合淘天开源电商经营评测基准1天前zg – Zvec AI 开源的本地优先统一检索基础工具1天前MAI-Transcribe-2 – 微软推出的最强 AI 语音转文字模型1天前GPT-6 Astra – OpenAI 推出的最新旗舰大模型1天前Cheso – 腾讯推出的 AI PPT Agent 制作工具2天前Muse Spark 1.3 – Meta 推出的新一代大语言模型2天前Easel – 浙大联合北大开源的 AI 社交媒体内容工作台2天前Fusion Model – FUMO Lab 推出的智能分配模型3天前Hy4 preview 轻量版 – 腾讯混元开源的Hy4量化压缩模型3天前Muse Voice Transcribe – Meta 推出的首个实时音频感知模型3天前Atlas – World Labs 推出的全球首个多模态世界模型3天前