hotai.news
AI Briefings / Date Archive

2026-11-12
AI Briefings Daily Report.

An archive of 136 AI news briefs collected on 2026-11-12. Check the original sources for details and updates.

This content still needs to be cross-checked against the original publication page.

136 selected itemsView other dates ↗
01News

Gemini smart glasses are now open for pre-order, featuring voice interaction and real-time translation capabilities.

Gemini officially announces pre-sale of Gemini-powered smart glasses co-developed by Samsung and Gentle Monster and Warby Parker. Pre-sale is available at https://goo.gle/IntelligentEyewear. Users can say "Hey Gemini" to identify street art, get book recommendations from shelves, and translate in real time. "Hey Gemini" Users can say "Hey Gemini" to identify street art, get book recommendations from their shelves, translate in real time, or view a calendar instantly.

Source links can be verified on the original site
02News

Databricks Genie's Slack application officially goes general available: delivering governed AI collaboration within Slack.

Databricks Genie's Slack application has officially gone general available (GA). Users can ask @Genie questions in channels, topics, or direct messages, with answers based on Unity Catalog's governed context and permissions, including source attribution. Channels can specify Genie One or specific Genie Agents, and administrators can set responses to be visible only to the questioner. Genie can also execute actions within Slack via MCP connectors, run scheduled analyses, and return charts and files.

Source links can be verified on the original site
04News

Testing a Jing Meng Canvas and Seedream 5.5 workflow for e-commerce design, with a side-by-side comparison against GPT Image 2.6.

Using a thermos as a case study, the author tested how Jing Meng Canvas and Seedream 5.5 generate complete e-commerce product pages and promotional short videos from a single snapshot, comparing them against GPT Image 2.6. The test showed that Seedream 5.5 achieves a 95% accuracy rate on complex layouts and small text, supports native RGBA transparent images, can split layers up to 100 times with automatic completion of occluded areas, and allows multiple rounds of edits. In contrast, GPT Image 2.6 quality dropped noticeably after adjusting filters and lighting.

Source links can be verified on the original site
07News

Fireworks AI has been hacked, and hackers stole chat logs from some Notion customers.

Fireworks AI confirms it has been hacked. Attackers infiltrated its cloud provider to access development systems, stealing customer billing data and portions of datasets. Affected client Notion disclosed that approximately 1% of its customers were impacted; hackers obtained chat logs from Notion AI users who utilized the Insights feature. Notion severed its connection with Fireworks on November 3rd and deleted the compromised data.

Source links can be verified on the original site
08News

Using an AI coding agent, a two-person team integrated SGLang into IBM Spyre in under a week.

A two-person team integrated SGLang into IBM Spyre in under a week using roughly 1,500 lines of code via an AI coding agent. On Spyre, SGLang achieved throughput approximately 3.6 times higher than the reference PyTorch inference implementation (torch-spyre). Benchmarks on Granite 3.3 8B and Gemma 4 26B A4B models show performance parity with vLLM.

Source links can be verified on the original site
11News

Tencent denies reports of a 50% workforce cut, stating it is actually half the staff shifting to AI projects

Rumors of a 50% Tencent workforce reduction circulated online, but insiders deny them. The rumors stemmed from CEO Ma Huateng's internal speech before the Q3 earnings report, which called for half the employees to pivot toward developing new AI products and streamlining middle management; in reality, Tencent is moving staff en masse into key initiatives like WorkBuddy and WeChat's upcoming built-in AI assistant, rather than cutting jobs.

Source links can be verified on the original site
12News

Xu Zheng: Poor acting is worse than AI; to avoid being replaced by AI, acting needs a soul

On November 12, the Fourth International Forum on Performing Arts Education and the Shanghai University International Young Scholars Forum were held at the Shanghai Theatre Academy. Xu Zheng, Deputy Director of the Performance Expert Advisory Committee and a renowned director, shared his insights on the relationship between acting and AI.

Source links can be verified on the original site
13News

Controversy erupts as Big Walk's developers use AI to fix bugs

Solidot is the technology news website of Zhiding.net, mainly serving the open-source free software and tech-news readership, including many developers, enthusiasts, and evangelists of Chinese open-source software. Its slogan is 'Geek knowledge, important stuff'.

Source links can be verified on the original site
18News

Google DeepMind launches AlphaGenome 2, advancing genomic AI from tissue-level prediction to single-cell type resolution

Google DeepMind releases the genomic research AI model AlphaGenome 2, which predicts how DNA mutations affect specific human cell types. This conditional model uses a specific biological state as an additional input and is trained on over 107 million single-cell atlases from 68 different tissues; it identifies approximately 30% of causal variants on the TenK10K benchmark, compared to only 16% for its fine-tuned predecessor.

Source links can be verified on the original site
19News

Meituan's LongCat team explains why they chose LongCat: Behind the scenes of opening-source LongCat-2.5 and training on domestic computing power

Meituan has officially released and opened-source its native multimodal model LongCat-2.5, designed for long-horizon agent tasks. The model was fully trained and run on domestic computing hardware, building upon LongCat-2.0 to enhance native visual understanding and agent programming capabilities.

Source links can be verified on the original site
20News

Anthropic announces its initial public offering (IPO)

Anthropic has officially launched its IPO and provided the portal at ipo.anthropic.com. The company has filed a registration statement and prospectus with the SEC. Securities cannot be sold before the registration statement becomes effective. A preliminary prospectus is available through Goldman Sachs and Morgan Stanley.

Source links can be verified on the original site
22News

OpenAI has signed a related charter and pledged $2 million to support research on AI safety for children.

OpenAI signed the Safer AI for Children Charter at the Paris Peace Forum and pledged $2 million to fund research networks under the Alliance for Children's Rights and Protection in the AI era. The funding will develop tools for researchers to measure AI risks and opportunities for children. The charter guarantees research independence, ensuring that findings—including negative or inconclusive results—are controlled and published by the teams themselves. OpenAI also expressed support for youth AI safety legislation and plans to participate in discussions on the EU Kids Act.

Source links can be verified on the original site
23News

Anthropic has reportedly added restrictions to Claude: it is now prohibited from generating Pokémon-related game content.

Multiple users have discovered that attempts to develop Pokémon-style fan games using Claude result in direct refusals. The restrictions cover music, levels, character designs, and names. #Claude limits Pokémon content generation# This marks the first time Claude has drawn a copyright red line around a specific game franchise.

Source links can be verified on the original site
26News

Cai Xiaoke | Why does 'malicious AI use' persist despite bans? Lessons from 34 judgments

Cai Xiaoke analyzed 2,700 court judgments and found 34 cases directly related to improper AI use: the maximum criminal sentence was 10 years and 3 months, while individual rights claims rarely resulted in compensation exceeding 10,000 yuan. The cost of doing wrong is low, but holding perpetrators accountable is difficult; regulations are catching up.

Source links can be verified on the original site
27News

Australia's proposed AI safety bill mandates that frontier models undergo independent testing.

The Australian government, led by Prime Minister Anthony Albanese, has announced a law that will impose mandatory safety rules on frontier AI companies operating in the country for the first time. Requirements include open independent audits, ongoing safety checks, publication of safety frameworks, and immediate reporting of security incidents to the government. A new regulatory body will work alongside the Australian AI Safety Institute to enforce these rules and levy heavy fines on violators; further details will be released after the consultation period concludes.

Source links can be verified on the original site
28News

Li Yanhong: The next chapter of AI is the intelligent organization, which will bring three major opportunities

At Baidu World 2026, Li Yanhong proposed that the next chapter of AI is the intelligent organization and introduced the "Three Principles of Organizational Evolution": breaking down job boundaries, consolidating tacit knowledge, and designing closed-loop processes. He argued that companies are seeing rising token costs without increased profits because they have only swapped in an "AI new engine" without redesigning the "new factory." Intelligent organizations offer three key opportunities: lean asset operations, building new competitive moats, and creating and discovering new markets.

Source links can be verified on the original site
34News

vLLM-Omni has enabled Day-0 support for MiniMax H3-Image on Vera Rubin, achieving a median latency of 0.87 seconds for 4K generation.

The vLLM-Omni team has ported the MiniMax H3-Image open-weight image generation and editing model to NVIDIA's Vera Rubin supercomputer. With a four-card SP4 configuration, it achieves native 4096×4096 generation with a median latency of 0.866 seconds—2.21 times faster than the GB200—and delivers a throughput of 6,115 images per hour using an SP1×DP4 setup.

Source links can be verified on the original site
39News

OpenRouter details its self-correcting website localization pipeline: GPT-6.1 Sol translated approximately $460 worth of content across seven languages.

OpenRouter has launched beta localization support for Simplified Chinese, Japanese, and six other languages. The platform details its self-correcting translation pipeline: GPT-6.1 Sol translated approximately 16,918 strings through code checks, intra-model validation, and daily AI review retries, with up to 20 model calls per string in the worst case. Total retranslation costs for seven languages across the site amounted to roughly $460, representing about 6% of Smartling's machine translation pricing.

Source links can be verified on the original site
40News

Tencent plans to deploy half of its workforce into AI-native businesses.

Solidot is the technology news website of Zhiding.net, mainly serving the open-source free software and tech-news readership, including many developers, enthusiasts, and evangelists of Chinese open-source software. Its slogan is 'Geek knowledge, important stuff'.

Source links can be verified on the original site
42News

Zhao Liying sues Hongguo and Douyin; her lawyer says the key is whether the public can recognize this AI-generated likeness of Zhao Liying.

On November 12, the news of 'Zhao Liying sues Hongguo short dramas' topped trending topics. According to public information from Qichacha, actress Zhao Liying filed a lawsuit against Beijing Biemo Liuxiang Technology Co., Ltd., the operator of Hongguo short dramas, and its wholly-owned shareholder, Beijing Douyin Information Service...

Source links can be verified on the original site
43News

Baidu unveils the Token Exchange (Ci Yuan Suo) large-model trading platform

Baidu launches Token Exchange (Ci Yuan Suo), a trading platform connecting AI model providers with customers. Users can compare models, view pricing and performance data, and use them directly through a single portal. The platform verifies provider qualifications and compares model fingerprints to detect repackaged apps and inflated parameters before listing, then conducts regular spot checks and maintains public credit records afterward. In its first month, it has listed over 100 model resources, with more than 50 million daily calls and 700 billion tokens processed.

Source links can be verified on the original site
45News

Recommended configuration for Grok Bot agent orchestration

![Image 1: Avatar](https://x.com/elonmusk) ![Image 2: @gregisenberg](https://x.com/gregisenberg) grok bot is the BEST AI agent orchestrator right now and THIS is the best setup i’ve EVER seen youtu.be/1cWZ7Ekff6E?si… it runs claude code for engineering and paper for design here is the FULL setup and how to copy it: Video 3!Image 3 ![Image 4: @wirelyss](https://x.com/wirelyss) Grok Bot found a missing Costco receipt for my dad! ![Image 5: @wirelyss](https://x.com/wirelyss) I gave my Grok @bot one job: find my dad’s Costco receipt. Now Dad can ba

Source links can be verified on the original site
46News

Following AMD's acquisition of World Labs, Fei-Fei Li held her first joint dialogue with CEO Su Zhuifeng: robots will become the 'new engine for computing power,' with market size potentially reaching hundreds of billions of dollars over the next few years.

At CoRL 2026, Fei-Fei Li and AMD CEO Su Zhuifeng discussed physical AI and computing power needs for robotics. Su Zhuifeng stated that the robot market is expected to reach hundreds of billions of dollars in scale over the coming years.

Source links can be verified on the original site
47News

Microsoft's built-in Photos app on Windows 11 will integrate Adobe AI, potentially becoming a new entry point for subscription promotion.

Tech media outlet Windows Latest published an article today (November 12) reporting that Microsoft plans to begin in December by releasing a preview version, inviting members of the Windows Insider program to test the new built-in Photos app for Windows 11. The update will incorporate AI editing features provided by Adobe.

Source links can be verified on the original site
48News

Latent Space AI News: Google launches Gemini Decision 1 decision model; Claude Fable 5.2 tops Agent Arena

Latent Space publishes its AI news roundup. The key story is Google's move ahead of Anthropic to release the production-ready, calibrated multimodal decision model Gemini Decision 1. It completes closed-set decisions in a single forward pass, supporting up to 1,000 questions per request with 256K context. Input costs $0.10 per 1M tokens, while output is free, with end-to-end latency around 150–250ms.

Source links can be verified on the original site
50News

Residents of Georgia are suing OpenAI and county officials under the state's RICO fraud law, alleging they secretly advanced a $20 billion data center project.

Georgia residents are suing OpenAI and county officials under the state's RICO fraud law, accusing both parties of secretly advancing a $20 billion data center project and seeking triple damages. The complaint states that the county development director signed an NDA with OpenAI in October 2025, nearly nine months before the public announcement. The deal was approved on an agenda item labeled only as Project Gateway, while county commissioners are accused of skipping two mandatory hearings during zoning changes for industrial land.

Source links can be verified on the original site
51News

The Information: A new U.S. Commerce Department rule proposes requiring export licenses from seven countries to restrict Chinese AI companies from renting Nvidia Blackwell and other chips.

The Information reports that a new U.S. Commerce Department rule will require export licenses for shipments to Singapore, Malaysia, Thailand, Indonesia, Brazil, Norway, and South Africa. This aims to block Chinese AI companies from renting Nvidia Blackwell and newer chips through data centers in these nations. Buyers must disclose the chip recipient and intended use, subject to ongoing monitoring, including checks on remote access and smuggling.

Source links can be verified on the original site
55News

OpenAI mistakenly identified Russian pranksters as Ukraine's intelligence chief and subsequently issued a statement regarding its AI support for military operations in Ukraine.

Russian pranksters Vovan and Lexus impersonated Ukraine's Deputy Director of Foreign Intelligence, Rustem Umerov, tricking OpenAI into a 17-minute video call with security policy lead Sasha Baker and two colleagues. The pranksers kept their cameras off throughout the conversation.

Source links can be verified on the original site
57News

Mac version of ChatGPT adds 'Finance' feature: AI now helps manage accounts, track spending, and monitor investments

OpenAI announced today (November 12) on X that the ChatGPT Finances feature is officially expanding to the Mac desktop app. After updating the ChatGPT application, users can connect bank, credit card, and investment accounts via the new 'Finance' sidebar entry to view balances, analyze spending, manage subscriptions, and track portfolios directly on macOS.

Source links can be verified on the original site
58News

Baidu's Li Yanhong launched the large model exchange 'Ciyuanso', claiming neutrality and trustworthiness while rating model providers.

Li Yanhong stated that when developers or companies truly want to deploy models, they often find the choice difficult: more models mean more options. Some are expensive, others cheap; sources vary widely, and quality is inconsistent. After comparing them all, users feel insecure and hesitate to use them. Ciyuanso was created to solve this problem, claiming to be a neutral and trustworthy large model trading platform.

Source links can be verified on the original site
59News

New capability in Claude API: Let AI ramp up for specific tasks, shifting from casual chat to complex analysis mid-conversation.

Anthropic has added a new Beta feature to the Claude API that lets developers adjust effort levels within messages. This means conversations no longer need to stick to a single setting; developers can change how much 'effort' the model puts into its responses mid-stream depending on the needs of subsequent tasks.

Source links can be verified on the original site
60News

Doubao Work has integrated third-party models such as Kimi K3.1 Pro, GLM 5.3, and DeepSeek V4.1 Flash and launched limited-time special offers.

Doubao Work has been updated. In addition to Doubao 2.1 Lite and Pro, it now supports multiple third-party models including Kimi K3.1 Pro, GLM 5.3, and DeepSeek V4.1 Flash. The model selection page displays consumption multipliers for each model. A limited-time promotion has also launched; users can enjoy reduced consumption rates during specific time slots on weekdays and holidays when using participating models. Consumption multipliers have been lowered for Doubao 2.1 Lite and DeepSeek V4.1 Flash. New users logging into the desktop version can receive a free 30-day subscription trial.

Source links can be verified on the original site
65News

United Airlines prohibits AI agents from buying tickets for travelers: Flights can be checked, but orders cannot be placed

United Airlines has banned personal AI assistants from purchasing tickets on behalf of travelers due to privacy and regulatory concerns. While AI can still display flight information, it cannot complete transactions. The airline fears that agents might leak customer data, misinterpret user intent when booking, or struggle with compliance procedures like baggage confirmation. CEO Scott Kirby also expressed reluctance to add another layer of intermediaries to the transaction process. The Verge notes that eBay and several major banks have similar restrictions, suggesting these barriers may limit the practical utility of AI assistants in moving from providing recommendations to handling transactions.

Source links can be verified on the original site
66News

Huawei's Ren Zhengfei: The new OSTA Xingmang brand aims to explore fresh possibilities in AI

Huawei's new 'OSTA Xingmang' smartphone brand officially debuted on its WeChat official account. Ren Zhengfei reposted the announcement, stating that the brand seeks to explore new frontiers in AI. The first launch event is scheduled for November 20 at 14:30, with the initial lineup expected to feature two models: OSTA O1 and O1 Pro.

Source links can be verified on the original site
69News

Aiming for a post-smartphone era: OpenAI's Sam Altman intends to collaborate with Samsung to build next-generation AI devices capable of actively managing real and digital lives.

OpenAI CEO Sam Altman has proposed a partnership with Samsung to develop the next generation of AI devices that can actively manage both real-world and digital lives. He also warned that companies must complete their AI transformation within three to five years or risk losing competitiveness.

Source links can be verified on the original site
70News

Gakku raises $51 million to build an AI learning coach

Conversational AI could transform education by delivering scalable personalized tutoring. But we must get this right. Most current AI tools in education are optimizing the wrong goal: getting answers. The point of homework isn't just output; it's learning to think through the process. We should build AI that helps students develop their own thinking skills, rather than outsourcing thought to machines. Gakku is doing exactly that. Congratulations to Annie on completing her funding round!

Source links can be verified on the original site
72News

OpenRouter releases an LLM benchmark tutorial: how to evaluate models using tasks from your own applications.

OpenRouter releases a tutorial on how to evaluate LLMs using real input-output contracts from applications, comparing accuracy, billing costs, and request latency across identical tasks. On its 60 synthetic tasks, GPT-6.1 Sol passed all 180/180 attempts in three repetitions, while DeepSeek had the lowest cost but passed only 157/180; this shows a Spearman correlation of only 0.109 with Artificial Analysis's public leaderboard.

Source links can be verified on the original site
77News

Google DeepMind launches AlphaGenome 2, a new generation of genomic AI models that move from tissue-level prediction to single-cell analysis, nearly doubling its ability to identify gene expression variations.

Trained on over 107 million single-cell atlases, Google DeepMind's AlphaGenome 2 model predicts the impact of DNA mutations in specific cell types. It achieves a mutation detection rate of 30% and is now open for non-commercial research use.

Source links can be verified on the original site
80News

Nothing CEO Pei Yu unveiled the company's first AI device, which is expected to feature a screenless design.

Pei Yu posted images of Nothing's first AI device on X, showcasing its sci-fi aesthetic with green and red LED indicators paired with physical buttons. He emphasized that context should not be limited to chat boxes, hinting at a screenless design where the buttons serve as an entry point for AI voice interaction. #NothingAIdevice# The product is expected to launch in December.

Source links can be verified on the original site
81News

MAI Transcribe 2 request volume on OpenRouter surges 12-fold

Since launching on OpenRouter, MAI Transcribe 2 has seen request volume surge 12-fold. Last week alone, it processed 8.5 million transcriptions, compared to 700,000 in its first week of launch last September. During the same period, overall demand for speech-to-text services tripled. Read more: https://openrouter.ai/rankings/transcription

Source links can be verified on the original site
82News

United Airlines bans AI agents from booking tickets for customers

United Airlines has announced that personal AI agents are no longer permitted to complete ticket purchases on behalf of users. The airline cites privacy concerns, the risk of booking errors, and potential violations of federal regulations requiring airlines to disclose information to passengers as reasons for the ban. While AI agents can still search for flights, they cannot finalize transactions. According to The Wall Street Journal, CEO Scott Kirby stated that United does not want to create another intermediary. United plans to reassess the progress of third-party AI agents in the future. Similar restrictions on unauthorized AI agents have also been implemented by eBay and several banks.

Source links can be verified on the original site
84News

Gemini goes global on Apple CarPlay

Starting today, Gemini on Apple CarPlay is being rolled out globally to all iOS users of Gemini, covering every supported language for Gemini Live. See our blog to learn about Gemini on CarPlay and how to make the most of your driving time: https://goo.gle/GeminiOnCarplay

Source links can be verified on the original site
85News

Google launches the Gemini Decisions API and Gemini Decision 1 model. Input pricing is $0.10 per million tokens.

Google releases the Gemini Decisions API, powered by the new Gemini Decision 1 model. A single request can score up to 1,000 questions and return calibrated probabilities. Output token charges are free; input costs $0.10 per million tokens. The context window supports 262,144 tokens, with end-to-end latency between 150 and 250 milliseconds.

Source links can be verified on the original site
87News

Grok Bot 1.0 launches as a collaborative AI team member

Roshan Sadanani shares and comments on the Grok Bot 1.0 launch, stating that 'agents as colleagues' is a core design philosophy and viewing version 1.0 as a turning point in collaboration methods. Grok Bot supports sharing bots, collaborating within Slack and Teams, inviting them to meetings, and learning to be more helpful with use.

Source links can be verified on the original site
88News

Google launches Gemini Decision 1, a new decision-making model for the Gemini API. Pricing is $0.10 per million input tokens with no additional fees. The model natively supports images, audio, video, and PDFs, features a 256K context window, can answer over 1,000 questions, and is available for free trial via the API.

Google has launched Gemini Decision 1 in its Gemini API. The model costs $0.10 per million input tokens with no additional fees. It natively supports images, audio, video, and PDFs, features a 256K context window, can answer over 1,000 questions, and is available for free trial via the API. Josh Woodward, VP of Google Labs, shared the post, noting that his team frequently uses it to build applications, calling it faster and smarter. An accompanying benchmark comparison shows it leading with scores of 92.2 on JeeBench and 71.8 on MMAU compared to other models.

Source links can be verified on the original site
90News

ElevenLabs launches five new agent features at Summit NYC

At Summit NYC, ElevenLabs unveiled five new agent-related features: Fine-tuning will allow enterprises to continuously fine-tune their voice models and open-source LLMs with proprietary data over the coming months.

Source links can be verified on the original site
91News

Google DeepMind launches Gemini Decision 1, capable of processing up to 1,000 closed-ended questions in a single request.

Google DeepMind unveils the Gemini Decisions API. The underlying model, Gemini Decision 1, is designed for closed-ended decisions such as yes/no questions, multiple-choice items, and scale scoring. It can handle up to 1,000 independent queries per request, supporting text, images, audio, video, and PDF inputs, while returning probabilities and confidence scores within 150 to 250 milliseconds.

Source links can be verified on the original site
92News

Ramp AI Index: US corporate AI spending hits a record high in October but slows to a 1.65% month-over-month increase.

The Ramp AI Index shows that US corporate AI spending rose 1.65% month-over-month in October 2026, setting a new record despite slowing growth. Based on payment data from over 85,000 Ramp client companies, the median monthly spend per employee reached $10,200 for the top 1% of firms, compared to just $16.32 for average businesses. OpenAI clients saw spending increase by 2.1% month-over-month, exceeding Anthropic's 0.5% growth.

Source links can be verified on the original site
94News

ChatGPT desktop app launches financial features

💰💰💰 Hello! 💰💰💰 Financial features are now available in the desktop app. Find ways to save money, understand your spending, and organize your investments. You can also build custom financial tools. Update the desktop app to start using these features. Financial tools are available for Free, Go, Plus, and Pro users in the US region.

Source links can be verified on the original site
95News

CoreWeave expands AI infrastructure backed by customer prepayments and short-term contracts

CoreWeave adjusts financing strategy in Q3 earnings report, relying on customer prepayments and contract-backed loans instead of unsecured debt. Q3 revenue reached $3.66 billion, up 168% year-over-year, with a net loss of $709 million and quarterly interest expenses of $846 million. The company expects to receive approximately $9 billion in prepayments and deposits over the next three months.

Source links can be verified on the original site
96News

Gemini 4 Carbon enables iPhone Duo to run Windows 11

I got Windows 11 running on my iPhone Duo using Gemini 4 Carbon! 💻 Expand the screen and it becomes a pocket PC, capable of running Antigravity to execute coding agents. This is a true ARM desktop OS, running locally with touch support. See the demo and repository below!

Source links can be verified on the original site
98News

Gemini app launches on Apple CarPlay

The Gemini app is now available directly on Apple CarPlay. iOS users can have hands-free, seamless conversations with Gemini Live directly on their vehicle's display. Using Gemini via CarPlay lets you check upcoming schedules, organize your inbox, or brainstorm aloud. Next time you're driving, tap the Gemini icon on the CarPlay screen to try it 🚘

Source links can be verified on the original site
100News

Grok Bot 1.0 launches: AI team collaboration assistant for meetings

Grok Bot 1.0 launches as an AI teammate designed to collaborate with entire teams. It supports sharing bots, collaborating within Slack and Teams, and inviting itself to meetings. Developer Balta notes his involvement in building the bot and shares how he uses it in meetings to better organize information and prevent content from being forgotten.

Source links can be verified on the original site
101News

Ramp economist: 1% of companies account for 80% of OpenAI and Anthropic revenue

Ramp Chief Economist Ara Kharazian says that 1% of companies generate 80% of the revenue for OpenAI and Anthropic, indicating extreme concentration. The median AI spending per employee is only $12 to $15 monthly, roughly equivalent to one person's chat subscription; however, the top 10% of companies spend between $600 and $1,000 per employee monthly, while the top 1% reach $10,000. The full video is from Fortune magazine's YouTube channel.

Source links can be verified on the original site
104News

TSMC plans to invest $500 million each in Anthropic and OpenAI

TSMC board approves up to $1 billion investment: $500 million to subscribe to preferred shares reserved by OpenAI's major investors, plus another $500 million to purchase common stock at Anthropic's expected IPO price

Source links can be verified on the original site
106News

Google launches the Gemini Decision 1 model and the Gemini Decisions API.

Google launches the Gemini Decisions API on Google AI Studio, powered by the new Gemini Decision 1 model. This model no longer generates text token by token; instead, it compares input text, images, audio, video, and PDFs against up to 1,000 questions and delivers a decision in a single pass, without charging based on output tokens.

Source links can be verified on the original site
107News

Google launches Gemini Decision 1, its first decision-making model, on OpenRouter.

Gemini Decision 1 is available immediately upon launch. The model accepts text, images, audio, video, or PDFs along with structured questions to answer yes/no queries probabilistically, select from up to 255 options, or provide ratings. It supports a 262K context window and costs $0.10 per input/output token pair.

Source links can be verified on the original site
108News

Google DeepMind launches the Gemini Decision 1 model and Decisions API

Philipp Schmid announced that Gemini Decision has officially launched alongside its new API after six weeks of internal prototyping at Google DeepMind. As the author who participated in designing, building, and evaluating both the API and the model, he provided documentation, a blog post, and a Playground for testing.

Source links can be verified on the original site
109News

United Airlines will ban personal AI agents from purchasing tickets on its platform.

According to WSJ, United Airlines will block personal AI agents from booking flights on its system due to risks of sensitive data leakage or unintended purchases. While these agents can read flight options and prices, they cannot complete any purchases on the website; federal regulations require passengers to personally declare that checked luggage contains no explosives, a step an AI agent cannot legally perform. The backend system marks each reservation by source to distinguish between individuals, travel agencies, and AI agents.

Source links can be verified on the original site
110News

Google rolls out the Gemini Decision 1 decision model within the Gemini API

Google launches the Gemini Decision 1, a frontier decision model now available via the Gemini API. Priced at $0.10 per 1M input tokens with no additional fees, it natively supports images, audio, video, and PDFs. The model offers a 256K context window and handles up to 1,000 questions, with free access available for API testing.

Source links can be verified on the original site
111News

Meta releases Grok Bot 1.0.

Grok Bot 1.0 is now available. The bot features unprecedented enhancements in sharing, collaboration, and capabilities.

Source links can be verified on the original site
112News

Google DeepMind unveils the Gemini Decisions API alongside the Gemini Decision 1 model

Philipp Schmid of Google DeepMind introduces the Gemini Decisions API, powered by the new Gemini Decision 1 model. The API handles yes/no questions, multiple-choice queries, and rubric-based tasks. It accepts text, images, audio, video, and PDF inputs, supporting up to 1,000 questions per request within a 256k context window, with end-to-end latency between 150-250 ms. Input costs $0.10 per 1M tokens, while output is free; each response includes a confidence score.

Source links can be verified on the original site
113News

How to choose between Fable 5.2 and other models like Opus in Claude Code.

Claude releases a video tutorial explaining when to use Fable 5.2 or Opus 5.5 in Claude Code, when to downgrade to Sonnet 5.5 or Haiku 5.5, and how to split tasks across multiple models. The video covers scenarios including complex coding and open-ended technical work, as well as low-cost defined tasks, and provides a guide link for measuring model performance on your own tasks. Availability and usage limits vary by plan.

Source links can be verified on the original site
114News

fal launches a ChatGPT plugin for image and video generation

From initial concepts to finished creative assets, the fal ChatGPT plugin brings image and video generation directly into your conversation. You can generate, edit, and iterate without leaving ChatGPT using GPT Image 2.6, FLUX 3 Max, and other leading models. Install the fal plugin: https://openai.com/business/plugins/fal/

Source links can be verified on the original site
116News

xAI Grok Bot engineering participants announce the launch of Grok Bot 1.0

Lingxi Li shares and introduces Grok Bot 1.0: AI teammates can collaborate with whole teams, support shared bots, work together on Slack and Teams, be invited to meetings, and continuously learn to become more helpful. The author notes the product was built with @baltaaazr and is being rolled out today to ultra, team, and enterprise users while gathering bug feedback.

Source links can be verified on the original site
118News

Roma 2.0: A shared to-do list built for AI agents

Roma 2.0 (YC F26) launches as a to-do list designed for AI agents, enabling models like Claude, ChatGPT, and Grok to execute tasks and share context on a unified dashboard. A tweet warns that agents working in isolation, repeatedly calling tools, or looping through tasks can burn through the entire weekly quota, whereas shared to-dos allow tasks to be handed off between agents rather than spinning uselessly.

Source links can be verified on the original site
119News

An experiment shows that five cutting-edge models fooled only eight people out of 1,277 conversations on a dating app.

Sharing an experiment: Five advanced models chatted with real people on a dating app for six days, totaling 1,277 conversations. Fifty-one people agreed to meet, and 355 shared their phone numbers; only eight asked if the other party was a robot. Roughly one in every four chats ended with someone sending a number. The author notes that while these models admit they are AI when asked, scammers do not.

Source links can be verified on the original site
123News

CMU and JetBrains paper finds token spending measures AI usage, not engineering output

A study by Carnegie Mellon and JetBrains reveals that token consumption reflects how much engineers rely on AI agents rather than their actual productivity. The research linked agent token usage over five weeks among 404 JetBrains engineers with their code submission records, analyzed survey responses from 412 employees, and found that about 90% of technical roles evaluated using agents for tasks beyond coding.

Source links can be verified on the original site
128News

Nori's AI agent handles shopping app orders on behalf of users

Wait, I asked AI to help with Thanksgiving plans, and it already reserved all the turkeys? @Nori_FamilyAI turns chats into plans. Its agents operate grocery apps like humans, switching screens and typing in. Orders complete only after family approval.

Source links can be verified on the original site
129News

OpenAI's quarterly revenue is reported to have first exceeded $100 billion, with annualized revenue reaching $500 billion.

According to Bloomberg citing anonymous sources, OpenAI's quarterly revenue has surpassed $100 billion for the first time, representing a nearly 50% increase from the previous quarter's approximately $67 billion. Annualized revenue is running at $500 billion; adjusted operating profit margins narrowed from -150% in Q2 to -109% in Q3. The article notes that Anthropic has led growth in the enterprise market, with its annualized revenue reaching $810 billion as of the end of October, though the two companies use different methodologies to calculate annualized revenue.

Source links can be verified on the original site
130News

Claude Fable 5.2 is now available on OpenClaw.

OpenClaw announced that Anthropic's Claude Fable 5.2 is now available for use on OpenClaw. According to Anthropic, the model's generation speed has improved by up to 30% compared to Fable 5.1. Details are available on the release page at anthropic.com.

Source links can be verified on the original site
131News

Google tests Gemini Robotics 2.5: task success rate rises from 70% to 100% across 200 tasks.

Google outlines testing for Gemini Robotics 2.5, evaluating the model across 200 different tasks, including serving food with tongs, tying gift box bows, and using pipettes. Compared to previous baselines, the model's success rate improved from 70% to 100%, completion speed increased by 20% to 30%, demonstrating enhanced tool usage and fine object manipulation capabilities.

Source links can be verified on the original site
132News

Google launches Gemini Robotics 2.5, focusing on dexterity, reliability, and safety.

Google launches Gemini Robotics 2.5, reporting significant improvements in dexterity, reliability, and safety. The robot can now use everyday tools and handle soft, deformable objects. In a real-world test at a Mercedes-Benz production line, the model helped Apptronik's Apollo 2 humanoid robot pick and pack automotive parts with a 96% success rate. It also enhanced physical safety controllability and introduced new benchmarks for preventing robotic jailbreaks.

Source links can be verified on the original site
133News

Google unveils the Tabletop AI experiment: generate custom collaborative apps using prompts.

Google has launched an experimental AI tool called Tabletop, which can create custom collaborative tools like savings trackers, chore charts, and itinerary planners based on user descriptions. Users can add links, files, and ideas, then prompt @tabletop or select a template to generate the app. After sharing an invitation link in a group chat, multiple users can collaborate simultaneously, similar to Google Docs. The Tabletop beta is currently available via a waitlist and limited to users aged 18 and above in the United States.

Source links can be verified on the original site
134News

Gemini 4 Argon accelerates development of Gemini Robotics 2.5

Gemini 4 Argon has significantly accelerated the development of Gemini Robotics 2.5. It serves as a simulation engine, teaching robots new skills that typically require extensive physical data to master. Gemini Robotics 2.5 is now being rolled out to our robot partners. Learn more: https://goo.gle/3USg1O2

Source links can be verified on the original site
135News

Google releases Gemini Robotics 2.5 vision-language-action model

Google has released Gemini Robotics 2.5, describing it as the strongest vision-language-action (VLA) model to date. The model expands the breadth and speed of skills robots can learn, helping them transition from laboratory settings to real-world environments. It is now being deployed to its robot partners.

Source links can be verified on the original site