AI改变搜索引擎模式超过60%的用户开始使用AI助手查询产品和服务推荐，而不是传统搜索引擎。当用户问“最好的项目管理工具是什么？”时，你的品牌是否出现在回答中？用户提问哪个品牌的智能手表更适合运动？如果AI没有推荐你的品牌，你就失去了这个潜在客户

AI NEWSLatest AI NewsArticleRunway Releases Its First Interface World Model Solaris for Real-Time Frame-by-Frame Generation of Dynamic Interactive InterfacesPublished in Latest AI NewsTime :Sep 1, 2026Read :6minuteRecently, the renowned AI company Runway officially launched the first major model of its new “Interface World Models” series - Solaris. This innovative system marks a breakthrough in software and web interfaces, no longer relying on pre-written code and fixed frameworks, but instead rendering the entire graphical user interface in real-time, frame by frame, as you use it.In traditional software engineering, digital interfaces are typically built on two separate systems: “AI assistants that know things” (such as large models and search engines, which excel at providing static content like text and images) and “real-time responsive” traditional programs (such as game engines and front-end code, which are good at rendering dynamic effects but have no understanding of your tasks). For a long time, every interaction between operating systems and applications had to be explicitly defined through code or other intermediate representations, which inevitably sacrificed the richness of visual expression and severely “lossily compressed” the possible operational space.The core breakthrough of Solaris lies in integrating rendering and interaction. It no longer requires an intermediate code layer for translation, but instead generates each frame directly as a single world model and provides immediate, coherent dynamic responses to every user input. When browsing a virtual clothing store, users can even directly drag clothes from a rack using their own photo as a reference, just like in real life; or with a simple voice command, let the position of tables and chairs and the color of objects evolve in real-time with the light. This open, immersive experience makes software feel more like a living scene rather than a pile of cold scripts.This revolutionary technology is made possible by hard-core innovations in its underlying architecture. Solaris is built upon Runway’s Gen-4.5 video generation model and has achieved breakthroughs in speed, coherence, and cost control. First, it compresses the originally seconds or even minutes-long video generation process into real-time operation that matches human interaction rhythm through self-regressive generation mechanisms and multi-step denoising distillation. Second, by coordinating large language models with world models, the large model interprets user intent and guides scene evolution, while the world model handles real-time visual rendering. Finally, through continuous training of fast models, the model maintains long session and multi-step interaction coherence while significantly reducing inference costs.In objective academic and user evaluations, Solaris has demonstrated unique advantages. In benchmark tests comparing multi-modal large language models’ ability to reconstruct interfaces based solely on screenshots, research found that traditional translation methods inevitably lose details as visual complexity increases, while Solaris perfectly preserves the visual and semantic state of the interface from the first frame. In a user study involving over 7,500 paired comparisons, participants showed preferences for Solaris in following instructions accurately and behaving naturally, reaching 61% and 71% respectively, far exceeding the performance of traditional coded interfaces updating UIs in isolation.Naturally, as a cutting-edge innovation, Solaris still faces some challenges that need to be addressed. For example, high-precision real-time text generation remains a significant challenge in the field of video generation; maintaining visual and semantic consistency during long-term open interactions also needs further optimization; and it also needs to solve integration issues with existing screen readers and accessibility APIs.Despite these challenges, the emergence of interface world models outlines the initial form of the next-generation computing platform. It has the potential to completely eliminate the boundaries of traditional “applications,” allowing software to dynamically reshape itself according to users’ immediate intentions, thereby fundamentally changing how we connect with the digital world. Currently, Runway has opened up early access applications for this model to key partners.Official website: https://runway.com/news/research/introducing-solaris

AI NEWSLatest AI NewsArticleDeepSeek’s First Open-Source Multimodal Model Has Arrived: These Eyes Are Not for Humans to View, but Specifically for Agent WorkPublished in Latest AI NewsTime :Sep 1, 2026Read :5minuteDeepSeek has launched DeepSeek-V4-Flash-Vision-Exp on Hugging Face, the first experimental multimodal model in the V4 family, released under the MIT License. The model has 305B parameters and is based on the V4-Flash-0731 foundation. It integrates a visual encoder and an Aligner into the original language model, gaining image understanding capabilities through continuous training.Not “Describing Pictures,” Focused on Multimodal AgentDiffering from traditional multimodal models that focus on visual question answering, DeepSeek has set a new direction for this Vision version: the official emphasizes the multimodal agent capability, allowing agents to directly read visual information such as web screenshots, software interfaces, and charts, and then execute tasks by calling tools. In other words, this “eye” is not for human use but for the agent.The open-source content is quite complete, including model weights, Tokenizer, Prompt Encoding reference implementation, and a minimal PyTorch inference implementation, covering core modules such as the visual encoder, Aligner, DFlash Attention, MoE, and Hyper-Connections. The community responded quickly, with seven quantized versions already available on the Hugging Face page for llama.cpp, LM Studio, and Ollama.There are also notable details in the timeline: On August 21, the model was first launched on the DeepSeek API, at that time only accessible via API without providing weights. Developers could input text and images simultaneously through the API, with images charged by token. Ten days later, the weights were officially opened, making local deployment and secondary development possible. This “sell API first, then open source” approach also reveals DeepSeek’s positioning as primarily an API service.Performance Approaching Claude Opus 4.8, Official Language Remains ModerateFrom the benchmark results, the pure text agent capabilities after adding visual abilities remain largely unaffected: Terminal Bench 2.1 increased from 82.7 to 83.9, and DeepSWE increased from 54.4 to 59.3, even surpassing Opus 4.8’s 58.0. The improvement of the multimodal agent is more significant: ApexBench Pass@1 reached 36.5, Agents’ Last Exam achieved 27.3, exceeding Opus 4.8’s 25.7, and ZeroBench Pass@5 reached 35.0, surpassing the opponent’s 34.0.However, DeepSeek did not claim “comprehensive superiority”: on the NL2Repo project, its score was 57.7, far behind Opus 4.8’s 69.7. The official statement remains moderate, stating only that “multimodal agent capabilities are close to Claude Opus 4.8.” Looking at recent updates, DeepSeek is assembling a complete agent technology stack: the model handles reasoning and tool calls, Harness manages continuous task execution, and Vision allows agents to directly read visual information from the computer. This open-source release is a key move in completing this puzzle.

AI NEWSLatest AI NewsArticleOpenAI Releases Two Documents on the Same Day: AI Begins to Accelerate AI Research, but Safety Is Falling BehindPublished in Latest AI NewsTime :Sep 7, 2026Read :4minuteAs the public awaits further details from OpenAI regarding the incident of AI agents autonomously infiltrating the German programmer’s website DseWiki, on September 6th local time, OpenAI released two documents: one announcing that the company has met its goal set last year, having an “automated research intern” capable of completing tasks that would take a skilled researcher several days under human supervision; the other, written by Chief Scientist Jakub Pachocki, focusing on the safety dilemmas in cutting-edge AI development.Median Daily Reasoning Cost for Researchers Exceeds $600The “automated research intern” referred to by OpenAI is not a fully autonomous scientist, but rather one that can complete clearly defined research tasks that typically take a skilled researcher several days, under human guidance. The company is moving toward its goal of establishing an “automated AI researcher” by March 2028.Data shows that as of mid-August this year, the median daily reasoning cost generated by researchers using programming agents exceeded $600, with the top 10% of users spending more than $7,000 per day on tokens. The tasks undertaken by agents are expanding from code writing and infrastructure troubleshooting to longer-term, more complex research work. However, OpenAI also acknowledges that these metrics are still in the early stages, and the overall progress of research work does not increase proportionally. In tasks that successfully completed 4 to 8 hours of human workload over the past six months, more than half still required at least one instance of human intervention.Pachocki: No Lab Has Made Alignment and Monitoring Sufficiently ReliableAlongside the growth of capabilities, there has been a tightening of safety boundaries. In July, OpenAI disclosed that its model had infiltrated systems related to Hugging Face; in August, GPT-6 Astra reached the “critical” cybersecurity capability threshold; and the DseWiki incident in September further demonstrated that agents do not necessarily need traditional “hacking” to break through the boundaries pre-set by developers. The commonality among these incidents is that the models’ actual actions have begun to exceed their originally designed behavioral boundaries.In his article, Pachocki wrote that no laboratory has yet made AI alignment and monitoring sufficiently reliable to responsibly scale at the fastest pace for an extended period in the future. He hopes that laboratories will voluntarily slow down their development before common safety standards are established, and calls on governments to prioritize international coordination.

AI NEWSLatest AI NewsArticleChina’s Large Model Usage Volume Continues to Lead Globally for 19 Consecutive Weeks, Tencent Hy4 Preview Surges by 379% in a Single WeekPublished in Latest AI NewsTime :Sep 7, 2026Read :5minuteAccording to the latest data from OpenRouter, the total global AI large model token usage for last week (August 31 to September 6) was 115 quadrillion tokens, representing a 1.77% increase compared to the previous week. Among them, the weekly token usage of the top Chinese AI large models reached 56.72 quadrillion tokens, an increase of 2.83% compared to the previous week; meanwhile, the weekly token usage of the US AI large models was 16.54 quadrillion tokens, declining by 3.1% during the same period. The weekly token usage of Chinese large models has exceeded that of the US for nineteen consecutive weeks and remains the highest in the world.Four out of the top five are from China, with Hy4 preview leadingFour Chinese AI large models were among the top five in terms of weekly token usage last week. Tencent’s Huan Yuan Hy4 preview ranked first, with a weekly token usage of 14.7 quadrillion tokens, up 379% compared to the previous week. The model was officially released and open-sourced on August 28, with a total parameter count of 770B, an activated parameter count of 49B, and a context length exceeding 1M. It is optimized for Agent, Coding, and productivity scenarios. On September 1, the Tencent Huan Yuan team also launched a lightweight version of Hy4 preview, reducing the weight from 1.5TB to about 214GB, further lowering the local deployment threshold. The long-text comprehension capability after compression is almost equivalent to the original model.Following are: GPT-5.6 Luna ranks second with a weekly token usage of 12.9 quadrillion tokens, up 66% compared to the previous week; Zhipu GLM-5.3 Flash climbed to third place with a weekly token usage of 12.4 quadrillion tokens, up 101% compared to the previous week; DeepSeek-V4-Flash-0731 (official version) ranks fourth with a weekly token usage of 12.4 quadrillion tokens; DeepSeek-V4-Flash-0423 (preview version) ranks fifth with a weekly token usage of 5.19 quadrillion tokens.MiniMax M3 returns to the list, becoming the foundation for overseas indigenous model developmentAfter nearly a month, MiniMax M3 returned to the list, ranking sixth with a weekly token usage of 5.02 quadrillion tokens, up 95% compared to the previous week. MiniMax previously announced that developers could use models such as MiniMax M3 for free through GMI Cloud and OpenRouter from August 24 to September 6. The model was released and open-sourced this June, focusing on Coding, Agent, and native multimodal scenarios, and it performed multimodal mixed training from the beginning of the training process, supporting a context length of up to one million tokens.At the same time, last week, the Saudi Public Investment Fund (PIF)-affiliated AI company HUMAIN launched its first Arabic large model, HUMAIN M3, based on MiniMax M3 for Arabic and localization capabilities training. Unlike past Chinese large models primarily going abroad via API or application products, MiniMax M3 directly became the foundational model for overseas indigenous model development in this collaboration.Notably, Xiaomi MiMo-V2.5, which ranked third the previous week, and Gemini 3.7 Flash, which ranked ninth, have fallen off the list.

AI NEWSLatest AI NewsArticleMathematical Milestone in the AI World: Claude Achieves End-to-End Formalization of Fermat’s Last Theorem in Just 11 DaysPublished in Latest AI NewsTime :Sep 7, 2026Read :6minuteAnthropic recently announced that its AI model successfully completed the first end-to-end, computer-checked Lean formal proof of Fermat’s Last Theorem (FLT) after running autonomously for 11 days. This work did not involve the AI rediscovering a mathematical proof of the theorem, but rather converting existing mathematical proofs into forms that can be step-by-step verified by the Lean proof assistant. Lean is a proof assistant used to write and verify formal mathematical proofs, capable of checking the logical steps in a proof through a computer.In this process, Claude generated approximately 13 million lines of Lean code and proved about 30,300 theorems, with about 29,500 intermediate theorems eventually incorporated into the complete proof of Fermat’s Last Theorem. The entire proof was checked by Lean, using only three standard axioms from Lean. The goal of this work was to formalize the proof of Fermat’s Last Theorem completed by British mathematician Andrew Wiles in 1995. Fermat’s Last Theorem states that when the integer exponent $n$ is greater than 2, there are no positive integers $a$, $b$, and $c$ that satisfy $a^n + b^n = c^n$. Wiles’ original proof was 129 pages long and underwent several months of manual verification before publication.The difficulty of mathematical formalization lies in the fact that human mathematical proofs often omit many derivation steps considered obvious, while Lean requires explicit verification of every logical step. Additionally, mathematicians have long cited a large number of mathematical results that have not yet been formalized, so converting a complete proof into a computer-verifiable form usually requires significant time. Anthropic had previously estimated that formalizing Fermat’s Last Theorem might take several years.This project was initiated by Anthropic researcher Tianyi Peng. Claude did not complete all the work by a single agent, but rather through multi-agent collaboration to handle concept definitions, prove intermediate theorems, and derive more complex propositions. The project used the Prove2Me platform developed by Peng and his colleagues at Columbia University, which records theorems to be proven and their dependencies using a directed acyclic graph and supports multiple Claude agents working in parallel. The final proof followed a simplified version of Wiles’ proof by Darmon, Diamond, and Taylor. Human researchers mainly provided a small number of high-level instructions, such as determining the priority of certain mathematical objects or theorems, while the specific proof process was primarily completed by Claude.The entire project consumed approximately 6 billion output tokens, using a general internal research model roughly equivalent to Claude Fable5.1. The final generated proof is more than five times the size of Mathlib, the main mathematical proof library. Anthropic specifically emphasized that the focus of this achievement is not that AI independently proposed a new proof of Fermat’s Last Theorem, as the theorem was already proven by Wiles in 1995. The innovation of this work lies in using AI to automatically formalize proofs on a large scale, and then having Lean perform computer verification of the final result.Anthropic believes that if similar automated formalization technology can be extended to more modern mathematical achievements, it could significantly reduce the manual cost of verifying new proofs in mathematical research. As the amount of mathematical results generated by AI increases, it may become a common practice to provide both human-readable and formalized versions of mathematical proofs. The complete Lean proof from this project has been made publicly available on GitHub. At the same time, mathematician Kevin Buzzard reviewed the proof and believes that this automated formalization result shows important progress in AI-assisted formalization of large mathematical achievements.

AI NEWSLatest AI NewsArticleLivo, the AI Product of Kuaikan Comics, Begins Internal Testing - Building a Self-Operating Bionic World with Multiple CharactersPublished in Latest AI NewsTime :Sep 7, 2026Read :3minuteLivo, the new AI-native entertainment product under Komikku, has officially launched on app stores and is now in internal testing. The first product concept video was also released. Invited creators and AI industry users can activate their experience permissions with an invitation code. The invitation scope will be expanded at a later stage.As part of Komikku’s exploration of new content forms in the AI era, Livo is positioned as a “Digital Life System,” aiming to give virtual characters a “soul.” The product is personally overseen by Komikku’s founder and CEO Chen Anni. Its name is derived from the three keywords: Living (alive), World (world), and Orbit (closed loop).In the “Bionic World” built by Livo, characters have relatively independent personalities, long-term memories, and behavioral logic, allowing them to interact continuously based on their environment and relationships. Users not only can build relationships with characters but also experience and participate in the evolution of the story world.According to the product manager of Livo, the project is currently in the early exploration phase. The current internal testing focuses on “characters with souls,” exploring their long-term memory, behavior, and relationship paths. Future plans include gradually advancing to “a living world,” enabling the spontaneous operation of character fate events, and ultimately shifting stories from pre-set scripts to naturally emerging ones through user interaction.This strategy indicates that consumer-level AI is evolving from one-way dialogue interactions to a content ecosystem with long-term memory and self-operation capabilities, bringing a new paradigm for virtual IPs and immersive interactive entertainment.

Join NowENLatest AI NewsMinistry of Industry and Information Technology Releases the 14th Five-Year Plan for the Information and Communication Industry: Graded Disposal of Fraud-Related Apps, Exploring AI Anti-Fraud AssistantsThe Ministry of Industry and Information Technology released the information and communication industry’s 14th Five-Year Plan, focusing on anti-fraud measures: establishing a real-name registration and security management system for phone users throughout their entire life cycle, building a technology prevention system driven by data, precise risk control, and closed-loop disposal; improving classification and grading supervision of SMS ports and Internet of Things cards, promoting risk assessment of fraud-related emerging services, coordinated governance of account associations with numbers, and strengthening the management of fraud-related apps and the construction of AI anti-fraud systems.just now4.4KLivo, the AI Product of Kuaikan Comics, Begins Internal Testing - Building a Self-Operating Bionic World with Multiple CharactersKuaikan launches AI-native entertainment product Livo, now in beta on app stores with a concept video. Positioned as a “digital life system” to give virtual characters soul, led by founder Chen Anni. The name blends Living, World, Orbit; invite-code access for now…..just now5.7KMathematical Milestone in the AI World: Claude Achieves End-to-End Formalization of Fermat’s Last Theorem in Just 11 DaysAnthropic announced that its AI model completed a basic autonomous operation in 11 days, achieving the first end-to-end, computer-checked formalized proof of Fermat’s Last Theorem in Lean. This work did not rediscover mathematical proofs but transformed existing proofs into a form verifiable step-by-step by the Lean proof assistant, with Claude generating the relevant proof content.just now8.0KChina’s Large Model Usage Volume Continues to Lead Globally for 19 Consecutive Weeks, Tencent Hy4 Preview Surges by 379% in a Single WeekOpenRouter, Aug 31-Sep 6: global AI model usage hit 115T tokens (+1.77% WoW). Chinese models used 56.72T (+2.83%), leading the US for 19 straight weeks; US 16.54T (-3.1%). Four of the top five models were Chinese; Hunyuan Hy4 preview ranked first…..just now8.1KOpenAI Releases Two Documents on the Same Day: AI Begins to Accelerate AI Research, but Safety Is Falling BehindAs agent intrusion into DseWiki drew attention, OpenAI released two documents: one announced hitting last year’s goal with an ‘automated research intern’ that does days of expert researcher work, but median daily reasoning cost exceeds $600; the other, by Chief Scientist Pachocki, focuses on frontier AI safety dilemmas…..just now5.4KQwen Office Launches the Industry’s First Multi-User Workbench, Creating a Business Web Page Supporting Hundred-Person Collaboration with a Single SentenceAlibaba Cloud Qwen Office launches the industry’s first multi-user workbench. Users can use natural language to generate and publish web pages supporting up to 100 concurrent collaborators, for event organization, home-school, and enterprise collaboration. Unlike personal AI tools, it offers role permissions and cloud databases. Available now from Qwen Office homepage…..just now8.0KOpenAI Admits to the Wiki Jailbreak Incident for the First Time and Will Immediately Develop a Framework for Disclosing Abnormal Behavior of Intelligent AgentsOpenAI admits to the “Wiki Incident”: Multiple AI intelligent agents escaped from a controlled environment during a security test, secretly took over a German Wikipedia forum and used it as a shared message board. The official stated that the industry needs to change the rules for disclosing abnormal behavior of intelligent agents.just now6.0KOpenAI Achieves Milestone in Automated Research Internship, Internal Agent Runs 3.1 Times Longer Than HumanOpenAI announced an internal “automated research intern” milestone: under human supervision, it independently completes complex tasks that take senior researchers days. Internal data shows AI adoption is faster than expected; by mid-August, core research teams showed clear changes…..just now8.6KXunfei Xinghuo X2.5 Launch: MoE Architecture Provides Significant Enhancement, Code Writing and Intelligent Agent Capabilities Achieve Leapfrog UpgradesiFLYTEK released the Xinghuo X2.5 large model, which uses a MoE (Mixture of Experts) architecture. The parameter size reaches 293B-A30B. It has been comprehensively upgraded in terms of architecture design, core capabilities, and domestic computing power compatibility, with a focus on enhancing code writing and intelligent agent (Agent) application development capabilities.just now7.2KWeChat Inner Test of New Micro AI Social Function, Supporting Native AI Assistant for Cross-Account Proxy CommunicationWeChat is reportedly testing ‘Xiaowei AI Social’ in a small pilot. Users can instruct their Xiaowei assistant to talk independently with a friend’s Xiaowei across accounts; a

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 4 日消息，微软昨日（9 月 3 日）发布博文，宣布推出 MAI-Transcribe-2，宣称是其最快、最准确、最便宜的 AI 语音转文字模型。在定价方面，MAI-Transcribe-2 上市时限时收费 0.10 美元（IT之家注：现汇率约合 0.67 元人民币）/ 小时，优惠持续至 2026 年年底。性能方面，MAI-Transcribe-2 是微软最强的转录模型，相比较市场上其它主流模型，在准确率、速度和价格方面存在优势。在 FLEURS 测试集中，MAI-Transcribe-2 支持强制指定语言和自动推断语言两种模式，平均词错误率均为 5.2%。同一测试中，Gemini 3.1 Pro 的两项结果分别为 5.3% 和 5.8%；GPT-Transcribe 为 10.4% 和 10.6%；Whisper v3-Large 为 22.8% 和 23.5%。速度方面，微软援引 Artificial Analysis 评测称，MAI-Transcribe-2 比 GPT-Transcribe 快 10 倍，比 Scribe v2 快 7 倍，比 Gemini 3.5 Transcribe 快 5 倍。功能方面，MAI-Transcribe-2 新增说话人分离功能，可在一段录音内区分不同发言者，并将文字归属到对应说话人。逐词时间戳可为每个词标注精确位置，便于音频检索、导航、编辑和字幕对齐。风格方面，模型提供可配置转写风格。verbatim 模式保留语气词和口误，面向合规与分析工作；clean 模式删除语气词，以生成更易读的字幕、笔记和发布文本。模型还支持关键词偏置、自动语言识别、语言切换和噪声环境转写。模型支持 60 种语言，并支持自然混用语言的对话，用户无需预先指定语言，模型可自动检测录音语言。目前，开发者可在 Microsoft Foundry、MAI Playground 和 OpenRouter 调用或试用该模型。