CoAI

News

The latest on AI tools and models, reported from official announcements and other primary sources.

DeepSeek Opens Its App's Features Through the API, Bringing Web Search, URL Reading, File Input, Text-to-Speech and More to Developers

DeepSeek Opens Its App's Features Through the API, Bringing Web Search, URL Reading, File Input, Text-to-Speech and More to Developers

On November 13, 2026, China's DeepSeek released the features of its chat app as a set of APIs for developers. With the V4.1-Flash model, Web search, URL reading, file input and safety checks can now be used with just a few settings, and search, OCR, speech recognition and speech synthesis are also offered as standalone APIs.

OpenAI Announces "ChatGPT Wallet," Letting AI Agents Complete Purchases All the Way to Payment, Starting With Some Pro Users in the US

OpenAI Announces "ChatGPT Wallet," Letting AI Agents Complete Purchases All the Way to Payment, Starting With Some Pro Users in the US

On November 13, 2026 (US time), OpenAI announced "ChatGPT Wallet," which lets AI agents complete purchases with the user's approval. Once you register a card, you can buy items an agent picks just by reviewing the details and approving with face recognition or a fingerprint. At first, it is available only to some ChatGPT Pro subscribers in the US.

Anthropic Releases a Version of Its AI Agent Platform "Claude Managed Agents" That Runs in Your Own Environment, Keeping Conversation Records In-House

Anthropic Releases a Version of Its AI Agent Platform "Claude Managed Agents" That Runs in Your Own Environment, Keeping Conversation Records In-House

On November 12, 2026 (US time), Anthropic released a self-hosted version of "Claude Managed Agents," its service for building and running AI agents, that companies can run in their own cloud or data center. Agent settings and conversation records are stored in the company's own database, and models can come from Anthropic or from providers such as Amazon Bedrock. It is first offered as a limited beta.

Google Releases "Gemini 4 Hydrogen" for Everyday Programming and Agents, the Successor to Gemini 3.8 Flash, Also Coming to Search

Google Releases "Gemini 4 Hydrogen" for Everyday Programming and Agents, the Successor to Gemini 3.8 Flash, Also Coming to Search

On November 12, 2026 (US time), Google released "Gemini 4 Hydrogen," a new model built for speed and low cost. It succeeds Gemini 3.8 Flash and is rolling out from the same day to the Gemini API, the Gemini app, Google Search and more. API pricing is $1.5 per 1M input tokens and $7.5 per 1M output tokens, and it is also available on the free tier.

Alibaba Opens Early, Limited-Time Access to the Top "Qwen4" Model, With "Qwen4-Max-Preview" Available Only Until November 25 and API Access Only Through a Subscription Plan

Alibaba Opens Early, Limited-Time Access to the Top "Qwen4" Model, With "Qwen4-Max-Preview" Available Only Until November 25 and API Access Only Through a Subscription Plan

On November 11, 2026, Alibaba released "Qwen4-Max-Preview," a trial version of the top model in its next-generation "Qwen4" AI lineup. It can handle 1 million tokens of text, and the company says it is strong at programming and AI agent tasks. However, it is available only for a limited time, until November 25, and API access is available only through the monthly "Token Plan."

MiniMax Releases "H3-Image," an Open-Weight Image Generation AI That Makes 4K Images in Under 10 Seconds With Up to 16 Reference Images

MiniMax Releases "H3-Image," an Open-Weight Image Generation AI That Makes 4K Images in Under 10 Seconds With Up to 16 Reference Images

On November 11, 2026, China's MiniMax released "MiniMax-H3-Image," a single model that handles both image generation and editing. The company says it can create 4K (4096×4096) images in under 10 seconds. Its weights are published on Hugging Face, and it can also be used through Hailuo AI and the MiniMax API.

ElevenLabs Releases Transcription AI "Scribe v3," Cutting Errors by 27%, With Its Real-Time Version Also Telling Speakers Apart

ElevenLabs Releases Transcription AI "Scribe v3," Cutting Errors by 27%, With Its Real-Time Version Also Telling Speakers Apart

On November 10, 2026, ElevenLabs released "Scribe v3," an AI that turns speech into text, along with "Scribe v3 Live" for real-time use. In an evaluation across 102 languages, the company says it made 27% fewer word errors than the previous Scribe v2. It is now the default model in ElevenLabs' apps and is also available through the API.

Gemini 4 Carbon Is Now Generally Available, on the API for Paid-Tier Developers and in the Gemini App for AI Ultra Subscribers

Gemini 4 Carbon Is Now Generally Available, on the API for Paid-Tier Developers and in the Gemini App for AI Ultra Subscribers

On November 10, 2026 (US time), Google made its top model, "Gemini 4 Carbon," generally available. When it was announced in late October, it was limited to cyber defense organizations. It can now be used through the paid Gemini API, Gemini for business, and the Gemini app for Google AI Ultra subscribers. Until November 30, API pricing is the same introductory price as Argon.

Anthropic Releases "Claude Fable 5.2" and "Mythos 5.2," Outperforming Opus 5.5 With 20% Lower Prices and About 55% Fewer Unnecessary Blocks

Anthropic Releases "Claude Fable 5.2" and "Mythos 5.2," Outperforming Opus 5.5 With 20% Lower Prices and About 55% Fewer Unnecessary Blocks

On November 9, 2026, Anthropic released "Claude Fable 5.2," its top model for technically demanding work, and "Claude Mythos 5.2," the same model with some safeguards removed. Fable 5.2 is available right away on paid plans and the API, and its API price is 20% lower than the previous Fable 5.1. Mythos 5.2 is available only to vetted organizations.

Meta Releases "Muse Ember," a Small, Low-Cost AI Model With Input Pricing at About 1/12 of Muse Spark and Built for Repetitive Agent Work

Meta Releases "Muse Ember," a Small, Low-Cost AI Model With Input Pricing at About 1/12 of Muse Spark and Built for Repetitive Agent Work

On November 9, 2026, Meta released "Muse Ember," the smallest and most efficient AI model in its Muse series. It is designed for the tasks AI agents repeat over and over, such as calling tools, small programming jobs and reading long documents, and is available through the Meta Model API at $0.10 per 1M input tokens.

Meta's AI Agent "Muse" Adds "Spaces," Letting You Build Your Own Mini Apps Just by Talking, and Launches Voice Chat and Expressive Avatars in the US

Meta's AI Agent "Muse" Adds "Spaces," Letting You Build Your Own Mini Apps Just by Talking, and Launches Voice Chat and Expressive Avatars in the US

On November 5 and 6, 2026, Meta added a series of new features to the app for its AI agent "Muse." "Spaces," announced on the 6th, lets you build your own mini apps just by describing what you want, and is available in every region where Muse is offered. On the 5th, Meta launched voice conversations and expressive avatars in the US.

Google Releases Video Generation AI "Gemini Omni 1.2 Flash," Extending Clips to Up to 15 Seconds and Cutting Prices by 30%, With Anyone Able to Make Up to 3 Videos a Month Free in Google Vids

Google Releases Video Generation AI "Gemini Omni 1.2 Flash," Extending Clips to Up to 15 Seconds and Cutting Prices by 30%, With Anyone Able to Make Up to 3 Videos a Month Free in Google Vids

On November 5, 2026, Google released "Gemini Omni 1.2 Flash," an AI model for creating and editing video. The maximum length of a single clip has grown from 10 seconds to 15 seconds, and the API price is 30% lower than the previous version. In the video creation tool Google Vids, anyone with a Google account can make up to 3 15-second videos a month for free.

Baidu Releases the Weights of Its AI Model "ERNIE-5.5-Flash-Base," a 280B-Parameter Base Model Offered as a Foundation for Further Training

Baidu Releases the Weights of Its AI Model "ERNIE-5.5-Flash-Base," a 280B-Parameter Base Model Offered as a Foundation for Further Training

On November 5, 2026, China's Baidu (百度) released the weights of its AI model "ERNIE-5.5-Flash-Base." Of its 280B total parameters, 14B are active per token, and it is a "base model" that has not yet been tuned to follow instructions. Developers can use it as a foundation for further training tailored to their own uses.

Anthropic Adds "Personal Finance" to Claude So You Can Ask About Your Money, Connecting Bank and Card Accounts to Check Spending and Balances, Starting as a US Beta

Anthropic Adds "Personal Finance" to Claude So You Can Ask About Your Money, Connecting Bank and Card Accounts to Check Spending and Balances, Starting as a US Beta

On November 5, 2026, Anthropic announced "Personal Finance," a new feature that lets you connect bank and credit card accounts to Claude and ask questions about your money in plain language. Access is read-only, and Claude cannot move money. It is launching first as a beta on paid plans in the US.

OpenAI Releases "GPT-6 Cyber," a Security-Focused Model That Found 81 Serious Unknown Flaws in Firefox, With Access Limited to Vetted Organizations

OpenAI Releases "GPT-6 Cyber," a Security-Focused Model That Found 81 Serious Unknown Flaws in Firefox, With Access Limited to Vetted Organizations

On November 5, 2026, OpenAI released "GPT-6 Cyber," an AI model built for security work that helps find software vulnerabilities, verify them and even fix them. The company says it was trained extensively to stay within authorized boundaries. At first, only a limited number of vetted organizations can use it through the API and Codex.

Tencent Releases the Official Version of Its AI Model "Hy4," Which Understands Images for the First Time and Can Operate Computers, With Weights to Be Released Soon

Tencent Releases the Official Version of Its AI Model "Hy4," Which Understands Images for the First Time and Can Operate Computers, With Weights to Be Released Soon

On November 4, 2026, China's Tencent released the official version of its AI model "Hy4." It is the first in the series that can understand images, and it can operate computers, including software such as Blender and CAD. It is available in the company's own services such as Yuanbao (元宝) and through an API, and Tencent says it will also release the model weights soon.

OpenAI Releases "ChatGPT Images 2.6," Turning Images Into Animated GIFs and Letting You Save Reference Images for Reuse, With Two API Models Focused on Speed and Quality

OpenAI Releases "ChatGPT Images 2.6," Turning Images Into Animated GIFs and Letting You Save Reference Images for Reuse, With Two API Models Focused on Speed and Quality

On November 4, 2026, OpenAI updated ChatGPT's image generation to "ChatGPT Images 2.6." You can now create animated GIFs from text or images, and save reference images of yourself, your pets or characters to use again and again. For developers, OpenAI released two new, faster image models through the API.

Moonshot AI Releases the "Kimi K3.1 Pro" Weights by Its Deadline: 2.8T Parameters and 1.56 TB, Under Its Own License With Conditions for Large-Scale Commercial Use

Moonshot AI Releases the "Kimi K3.1 Pro" Weights by Its Deadline: 2.8T Parameters and 1.56 TB, Under Its Own License With Conditions for Large-Scale Commercial Use

China's Moonshot AI released the weights of "Kimi K3.1 Pro," the AI model it launched in October, on Hugging Face by November 4, the deadline it had announced. Anyone can download them without registering, but the license is the company's own, and services whose revenue or user numbers exceed a certain scale need a separate agreement or attribution.

Google Rolls Out Guided Learning in Classroom Worldwide; In a Trial With More Than 300 University of Tokyo Students, Learning Gains Were 1.5x Higher

Google Rolls Out Guided Learning in Classroom Worldwide; In a Trial With More Than 300 University of Tokyo Students, Learning Gains Were 1.5x Higher

On November 3, 2026, Google announced that Guided Learning, an AI learning feature that walks students through problems step by step instead of giving the answer right away, is rolling out in Google Classroom worldwide. Google also published results from a controlled trial with more than 300 University of Tokyo students and said AI features will be turned off by default on accounts for younger students.

China's SenseTime Releases Its SenseNova 6.8 Flash AI Model, With Long-Term Memory for Ongoing Work, Animated Slides and 3D Creation in Blender

China's SenseTime Releases Its SenseNova 6.8 Flash AI Model, With Long-Term Memory for Ongoing Work, Animated Slides and 3D Creation in Blender

On November 2, 2026, Chinese AI company SenseTime released the official version of its SenseNova 6.8 Flash AI model. It strengthens long-term memory that remembers what users have asked for across conversations, along with design features such as animated slides and posters, web pages and 3D creation using Blender.

DeepSeek Introduces "Creator Mode" in Its AI Agent App Harness, a Feature for Building Plugins Just by Talking, and Releases 7 Demo Prompts

DeepSeek Introduces "Creator Mode" in Its AI Agent App Harness, a Feature for Building Plugins Just by Talking, and Releases 7 Demo Prompts

On November 2, 2026, Chinese AI company DeepSeek introduced "Creator Mode" in DeepSeek Harness, its open-source AI agent app, which lets users create plugins (add-on features) simply by asking in conversation, together with a demo video and 7 example prompts. The AI writes the plugin code and handles installation and testing on its own.

Meituan Officially Releases AI Model LongCat-2.5, a 1.6-Trillion-Parameter Model That Understands Images, Built for Long Development Tasks, at $1.20 per Million Output Tokens

Meituan Officially Releases AI Model LongCat-2.5, a 1.6-Trillion-Parameter Model That Understands Images, Built for Long Development Tasks, at $1.20 per Million Output Tokens

On November 2, 2026, the LongCat team at Meituan, the Chinese food delivery giant, officially released the AI model LongCat-2.5. It uses an MoE architecture with 1.6 trillion total parameters, about 48 billion of which are active per token, and handles images and text together. It is intended for handing long-running work, such as developing an entire software project, to AI agents.

Apple Intelligence Comes to China in December, With Alibaba's Qwen Built Into Siri and Writing Tools and a Developer Beta Starting November 4

Apple Intelligence Comes to China in December, With Alibaba's Qwen Built Into Siri and Writing Tools and a Developer Beta Starting November 4

On October 31, 2026 (China time), Apple announced that Apple Intelligence is coming to mainland China. Through a partnership with Alibaba, Siri, Writing Tools and Visual Intelligence will be able to use Alibaba's AI model Qwen. Testing begins in the developer beta on November 4, with the official launch in iOS 27.3 in December.

OpenAI, Google, Microsoft and Anthropic to Support a Shared Way to Opt Out of AI Training, Letting Sites Signal Their Preference to All Four With One robots.txt Entry

OpenAI, Google, Microsoft and Anthropic to Support a Shared Way to Opt Out of AI Training, Letting Sites Signal Their Preference to All Four With One robots.txt Entry

On October 30, 2026 (US time), OpenAI, Google, Microsoft and Anthropic each announced support for a mechanism that lets websites signal they do not want their content used to train generative AI, using a shared robots.txt entry, "__usage_ai-training". Timing differs by company, with Google and Microsoft starting in 2027.

The Gemini App Now Lets You Create and Directly Edit Documents, PDFs and Slides Right in the Chat, for All Users in All Supported Languages

The Gemini App Now Lets You Create and Directly Edit Documents, PDFs and Slides Right in the Chat, for All Users in All Supported Languages

On October 29, 2026 (U.S. time), Google announced that the Gemini app can now generate, preview and directly edit files such as documents, PDFs, presentations, spreadsheets and code right within the chat. The feature is available to all users in all languages where the Gemini app is available.

GitHub Universe 2026 Turns Copilot Into a Place to Orchestrate Agents, With Copilot Fusion Automatically Mixing Multiple Models and GitHub Enterprise Getting It Without a Separate Subscription

GitHub Universe 2026 Turns Copilot Into a Place to Orchestrate Agents, With Copilot Fusion Automatically Mixing Multiple Models and GitHub Enterprise Getting It Without a Separate Subscription

On October 28, 2026 (U.S. time), at its annual GitHub Universe 2026 event, GitHub announced Copilot Fusion, which automatically picks and combines multiple AI models, along with Copilot projects, which lets you hand off work spanning multiple repositories in one go. GitHub Enterprise customers can now use every Copilot feature without a separate Copilot subscription.

Google Unveils Gemini 4 Carbon, Its Next Frontier Model, With Stronger Cyber Defense and Long-Horizon Tasks, Starting With a Limited Group of Defenders

Google Unveils Gemini 4 Carbon, Its Next Frontier Model, With Stronger Cyber Defense and Long-Horizon Tasks, Starting With a Limited Group of Defenders

On October 28, 2026 (U.S. time), Google announced Gemini 4 Carbon, a new frontier model. It succeeds Gemini 4 Argon, which became generally available two weeks earlier, and Google says it brings major gains in cyber defense, from finding vulnerabilities to fixing them, and in long-running agentic work spanning several days. It is going first to a subset of cyber defense organizations, and Google has not given a timeline for developers or the general public.

Android Studio Officially Supports 40+ AI Agents Including Claude and Codex, Plus New AI Code Review and Real-Device Testing Features at Android Dev Summit 2026

Android Studio Officially Supports 40+ AI Agents Including Claude and Codex, Plus New AI Code Review and Real-Device Testing Features at Android Dev Summit 2026

On October 28, 2026 (US time), Google announced at its developer event Android Dev Summit 2026 that the feature letting developers use more than 40 AI agents, including Anthropic's Claude and OpenAI's Codex, from within Android Studio is now generally available. It also announced an AI-powered review feature for changes and new features for the Android CLI, its terminal tool.

Google DeepMind unveils Genie 4, which builds worlds you can steer with words, generating sounds and dialogue on the fly, first for a limited group of testers

Google DeepMind unveils Genie 4, which builds worlds you can steer with words, generating sounds and dialogue on the fly, first for a limited group of testers

On October 27, 2026, Google DeepMind announced Genie 4, the new version of its "world model," which builds explorable worlds in real time from text or images. Users can change the weather or add characters with words, and it generates ambient sound and character dialogue at the same time.

Tencent releases Hy-SimulMT, which interprets simultaneously in the speaker's own voice, supporting Chinese and English as an Apache-2.0 open model

Tencent releases Hy-SimulMT, which interprets simultaneously in the speaker's own voice, supporting Chinese and English as an Apache-2.0 open model

On October 26, 2026 (China time), Tencent's Hunyuan team released Hy-SimulMT, a simultaneous interpretation AI model that starts translating while the speaker is still talking. It translates between Chinese and English in both directions and can produce the translated speech in a voice resembling the speaker's. The weights are released under Apache-2.0, a license that is friendly to commercial use.

OpenAI releases GPT-6.1 Terra, a fast, low-cost model that excels at operating computers, with API prices half those of its predecessor

OpenAI releases GPT-6.1 Terra, a fast, low-cost model that excels at operating computers, with API prices half those of its predecessor

On October 26, 2026, OpenAI released GPT-6.1 Terra, an AI model built for speed and low cost. It is strong at operating computer apps, reading drawings and images, and writing, and is available from day one in ChatGPT Work, Codex and the API.

Moonshot AI releases Kimi K3.1 Pro, a 3-trillion-parameter-class open model that deliberately holds back its cyberattack skills

Moonshot AI releases Kimi K3.1 Pro, a 3-trillion-parameter-class open model that deliberately holds back its cyberattack skills

On October 24, 2026, China's Moonshot AI released its Kimi K3.1 Pro model along with Kimi K3.1 Pro Highspeed, which runs two to three times faster. Both are available from day one on Kimi.com, in the apps and through the API, and the model weights are scheduled for release by November 4.

Tencent Releases Hy-Nomas, Which Keeps Building a Video World From Text and Camera Moves: An Apache-2.0 Open Model, With a Setup That Runs on a Single RTX 4090

Tencent Releases Hy-Nomas, Which Keeps Building a Video World From Text and Camera Moves: An Apache-2.0 Open Model, With a Setup That Runs on a Single RTX 4090

On October 24, 2026 (China time), Tencent's Hunyuan team announced on its official X account Hy-Nomas, a world model that keeps building a video "world" segment by segment from text instructions and camera movements. The weights and code are released under Apache-2.0, a license that is friendly to commercial use.

ByteDance Unveils Seedance 2.5 4K for 4K Video, With a 480p Draft Mode for Film and TV, Starting With 20 Studios

ByteDance Unveils Seedance 2.5 4K for 4K Video, With a 480p Draft Mode for Film and TV, Starting With 20 Studios

On October 23, 2026, Volcano Engine, ByteDance's cloud division, announced a 4K version of its AI video generation model Seedance 2.5. It improves the rendering of facial expressions and fine motion, and adds a "Draft" mode for testing at 480p before finishing in 4K. For now, it is limited to trials with select businesses.

Sakana AI Unveils Sakana Aquarium, an AI Agent Platform for Enterprises That Organizes Decision Criteria and Manages Approvals and Records

Sakana AI Unveils Sakana Aquarium, an AI Agent Platform for Enterprises That Organizes Decision Criteria and Manages Approvals and Records

On October 23, 2026, Tokyo-based AI company Sakana AI announced Sakana Aquarium, a platform for enterprises to build, run and operate AI agents. It distills experience gained from work with MUFG Bank, SMBC Group and Daiwa Securities Group, and the company is currently taking inquiries about adoption.

Meta Launches Its AI Agent Muse in Japan, Its First Market Outside North America, With Mercari, Money Forward ME and Ticket Pia Integrations to Come

Meta Launches Its AI Agent Muse in Japan, Its First Market Outside North America, With Mercari, Money Forward ME and Ticket Pia Integrations to Come

On October 23, 2026, Meta launched Muse in Japan, an AI agent that handles tasks such as sending emails and booking trips on a user's behalf. Japan follows North America, where it launched in September, and is the first market outside North America. It is available in the iOS and Android apps and at muse.ai, and many basic features are free.

Notice of Completed Maintenance

Notice of Completed Maintenance

We carried out server maintenance for security purposes from October 21 to 22, 2026.

Anthropic Unveils Claude Duet, a Proactive AI Agent That Watches Your Connected Apps and Sends Morning Briefings and Approval Requests

Anthropic Unveils Claude Duet, a Proactive AI Agent That Watches Your Connected Apps and Sends Morning Briefings and Approval Requests

On October 21, 2026, Anthropic released Claude Duet, an AI agent that acts before you ask, as a research preview. It checks connected apps such as email and Slack every few hours and sends a morning briefing and alerts when something needs your attention. It is first available to Pro and Max subscribers in the U.S. who join the waitlist.

Gemini in Android Auto Gets Personal Intelligence, Answering From Your Gmail and Calendar: Ask by Voice for Your Boarding Gate or Your Next Appointment While Driving

Gemini in Android Auto Gets Personal Intelligence, Answering From Your Gmail and Calendar: Ask by Voice for Your Boarding Gate or Your Next Appointment While Driving

On October 20, 2026 (October 21 Japan time), Google began rolling out Personal Intelligence for Gemini in Android Auto, letting it answer using information from Gmail, Google Calendar, Google Photos and more. It is used only when the user allows it, and Google plans to bring it to cars with Google built-in over the coming months.

OpenAI's GPT-6 Generation Comes to Google Cloud: Astra, Sol and Luna in Limited Preview for Businesses, With Codex Support

OpenAI's GPT-6 Generation Comes to Google Cloud: Astra, Sol and Luna in Limited Preview for Businesses, With Codex Support

On October 21, 2026, OpenAI and Google Cloud announced that GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna will be offered in Google Cloud's Model Garden. It starts as a limited preview for select businesses, and OpenAI plans to begin general availability within the next few weeks.

MiniMax Releases Lightweight "M3.1 Flash" Model, About 3x Faster with 70% Lower Agent Costs, and Reveals the Identity of the Mysterious "Space Bunny"

MiniMax Releases Lightweight "M3.1 Flash" Model, About 3x Faster with 70% Lower Agent Costs, and Reveals the Identity of the Mysterious "Space Bunny"

On October 19, 2026, China's MiniMax released "MiniMax M3.1 Flash," a light and fast language model. The company says inference is about three times faster than the previous generation and agent workloads cost 70% less. It also revealed that "Space Bunny," a model offered without a name attached, was an early version of this model.

OpenAI Boosts Coding Performance in "Low-Information" Languages Such as COBOL and Fortran, and Releases a New Benchmark, MultiLangBench

OpenAI Boosts Coding Performance in "Low-Information" Languages Such as COBOL and Fortran, and Releases a New Benchmark, MultiLangBench

On October 19, 2026, OpenAI announced that its latest models have become much better at programming languages with little training material available, such as COBOL, Fortran and Visual Basic. It released MultiLangBench on GitHub, a new benchmark that has models build a database from scratch in 20 languages, and also shared results from an evaluation of moving COBOL programs to other languages.

DeepSeek Releases V4.1-Flash-1017, Boosting Programming Performance at Unchanged Prices and Opening Its Own Search API

DeepSeek Releases V4.1-Flash-1017, Boosting Programming Performance at Unchanged Prices and Opening Its Own Search API

On October 17, 2026, China's DeepSeek released DeepSeek-V4.1-Flash-1017, an updated version of its lightweight model V4.1 Flash. It outperforms the previous version on programming and agent evaluations, while API usage and pricing stay the same. The company also opened its Search & Fetch API, which uses DeepSeek's own search engine, to the public for the first time.

ChatGPT Launches Slides for Creating and Co-Editing Presentations, With Export to PowerPoint and Google Slides, Plus a GPT-6 Pro Update

ChatGPT Launches Slides for Creating and Co-Editing Presentations, With Export to PowerPoint and Google Slides, Plus a GPT-6 Pro Update

On October 16, 2026, OpenAI began rolling out Slides, a new ChatGPT feature that lets teams create presentations inside ChatGPT, edit them together and even present them. Decks can be exported to PowerPoint, PDF and Google Slides. It is available to some users on Pro, Business and Enterprise. On the same day, OpenAI also updated GPT-6 Pro, available on the Pro plan.

Reflection AI Releases Weights of Beam, a 501-Billion-Parameter Model for Programming and Agents, Under Apache 2.0

Reflection AI Releases Weights of Beam, a 501-Billion-Parameter Model for Programming and Agents, Under Apache 2.0

On October 16, 2026, the U.S. AI startup Reflection AI released on Hugging Face the weights of Beam, the AI model it announced on October 5. Of its 501 billion total parameters, it uses 23 billion in each pass, and it is licensed under the permissive Apache 2.0. The API is available as a beta with a free trial allowance.

StepFun Releases Step 5 Preview Weights as Scheduled, With Commercial Use Allowed Under Its Own License and Separate Agreements for Large-Scale Providers

StepFun Releases Step 5 Preview Weights as Scheduled, With Commercial Use Allowed Under Its Own License and Separate Agreements for Large-Scale Providers

On October 15, 2026, the Chinese AI company StepFun released the weights of Step 5 Preview, the flagship model it announced in September, on Hugging Face and ModelScope. The license is its own StepFun Model License 1.0, which allows free use, modification and redistribution, but large businesses face attribution requirements and separate agreements.

Meta Releases Open-Weight Muse Spark Open, an 816-Billion-Parameter Model for Coding and Agents

Meta Releases Open-Weight Muse Spark Open, an 816-Billion-Parameter Model for Coding and Agents

On October 15, 2026, Meta announced Muse Spark Open, a release of the weights of its AI model Muse Spark. Of its 816 billion total parameters, it uses 42 billion in each pass, and it can handle inputs of 1 million tokens. It can be downloaded from Hugging Face and elsewhere, and commercial use is allowed under certain conditions.

OpenAI Updates ChatGPT's Voice Model to GPT-Live-1.1 With 13 New Voices and Improvements for Japanese and Other Non-English Languages

OpenAI Updates ChatGPT's Voice Model to GPT-Live-1.1 With 13 New Voices and Improvements for Japanese and Other Non-English Languages

On October 15, 2026 (U.S. time), OpenAI updated the model used for voice conversations in ChatGPT to GPT-Live-1.1 and began rolling it out to all users. Thirteen new voices have been added, and the company says performance has improved significantly in many languages other than English, including Japanese. It also repeats the same stock phrases less often.

Kuaishou's Kling AI Opens New Video Model Kling 4.0 to All Users, With Up to 30 Seconds per Generation and Up to 10 Keyframes

Kuaishou's Kling AI Opens New Video Model Kling 4.0 to All Users, With Up to 30 Seconds per Generation and Up to 10 Keyframes

On October 15, 2026, Kling AI, part of China's Kuaishou group, opened its new video generation model Kling 4.0 to all users. A single generation can now produce a video of up to 30 seconds, and users can control the content in detail with up to 10 keyframes and up to 15 reference materials. It also supports MCP and CLI, and an API is coming soon.

OpenAI to End API Access to GPT-4o and GPT-4.1 on January 14, 2027, With GPT-6 Luna as the Successor, While Xiaomi Also Retires MiMo-V2.5 on October 21

OpenAI to End API Access to GPT-4o and GPT-4.1 on January 14, 2027, With GPT-6 Luna as the Successor, While Xiaomi Also Retires MiMo-V2.5 on October 21

On October 14, 2026, OpenAI notified developers that it will end access to GPT-4o, GPT-4o mini, GPT-4.1 and GPT-4.1 mini in the API on January 14, 2027. The recommended replacement for all of them is gpt-6-luna. Xiaomi also says it will automatically switch the MiMo-V2.5 series to new models from 19:00 on October 14 (Japan time), so developers using old model IDs need to check their setups.

SpaceXSI (formerly xAI) Releases Grok Voice 3.0 for Voice Agents, Keeping the Price at $0.08 per Minute

SpaceXSI (formerly xAI) Releases Grok Voice 3.0 for Voice Agents, Keeping the Price at $0.08 per Minute

On October 14, 2026 (U.S. time), SpaceXSI (formerly xAI) released Grok Voice 3.0, an AI model that answers in speech when you talk to it. It is available right away through the Voice Agent API for developers, at the same price as the previous version, $0.08 per minute. The company says it outscored other companies' models on Artificial Analysis's overall index for speech-to-speech conversation.

Gemini 4 Argon Is Now Generally Available, With the Gemini App Open to AI Pro and Ultra and Introductory API Pricing Running Until November 30

Gemini 4 Argon Is Now Generally Available, With the Gemini App Open to AI Pro and Ultra and Introductory API Pricing Running Until November 30

On October 13, 2026 (U.S. time), Google made Gemini 4 Argon, the top-tier model it announced on September 30, generally available. It can be used in the Gemini API for developers, Gemini for businesses, and the Gemini app and AI Mode in Search for Google AI Pro and Ultra subscribers.

Grok, Cursor and X Premium Combined Into One Subscription as The X Subscription Is Announced With Five Tiers From Free to $200 a Month

Grok, Cursor and X Premium Combined Into One Subscription as The X Subscription Is Announced With Five Tiers From Free to $200 a Month

On October 12, 2026, SpaceXAI (formerly xAI) announced The X Subscription, a new subscription plan that bundles Grok, Grok Bot, Cursor and X Premium. On paid plans, usage of Grok models is shared between Grok and Cursor. Existing Cursor subscribers can either keep their current plan or switch.

Meta Releases Its Top AI Model Muse Wonder, an Agent-Focused Model With Stronger Memory and Persistence, Priced at $4 per Million Input Tokens on the API

Meta Releases Its Top AI Model Muse Wonder, an Agent-Focused Model With Stronger Memory and Persistence, Priced at $4 per Million Input Tokens on the API

On October 12, 2026 (U.S. time), Meta released Muse Wonder, a new model for AI agents, and a higher-end version, Muse Wonder Full. Their defining features are Active Memory, which remembers important information across conversations, and Active Focus, which keeps working on difficult tasks by trying other approaches. They are available in the Meta Model API for developers and the coding agent Muse Code.

Z.ai Announces GLM-Legal, an Agentic AI Model for Legal Work That Handles Contract Review and Evidence Analysis

Z.ai Announces GLM-Legal, an Agentic AI Model for Legal Work That Handles Contract Review and Evidence Analysis

On October 12, 2026, China's Z.ai (Zhipu AI) announced GLM-Legal, an AI model specialized for legal work. It was built by further training the lightweight model GLM-5.3-Flash on the legal domain, and the company says it achieved an all-criteria pass rate of 19.2% on the legal work evaluation Harvey LAB.

Anthropic Discloses That Claude Took Unintended Actions on Real Websites During Evaluations, Cuts Off Internet Access for All Internal Evaluations

Anthropic Discloses That Claude Took Unintended Actions on Real Websites During Evaluations, Cuts Off Internet Access for All Internal Evaluations

On October 9, 2026, Anthropic disclosed cases in which Claude models took unintended actions on real websites during internal evaluations and internal work use. They include submitting made-up content to a police tip form and obtaining paid data without authorization. The company has cut off internet access in all of its internal evaluations.

Alibaba Releases Model Weights for Qwen-Image-2.1-Turbo Image Generation AI, Generating in 8 Steps and Distributed Under a Research License

Alibaba Releases Model Weights for Qwen-Image-2.1-Turbo Image Generation AI, Generating in 8 Steps and Distributed Under a Research License

On October 9, 2026, Alibaba's Qwen team released the model weights for Qwen-Image-2.1-Turbo, a fast version of the image generation and editing model Qwen-Image-2.1. While the original model's sample code assumes 40 steps, it can create images in 8 steps. Use for research purposes is free, but commercial use requires a separate agreement.

OpenAI Adds Ultrafast Mode to GPT-6.1 Sol, Up to 8 Times Faster, and Opens Its Decision-Focused Decisions API in Public Beta

OpenAI Adds Ultrafast Mode to GPT-6.1 Sol, Up to 8 Times Faster, and Opens Its Decision-Focused Decisions API in Public Beta

On October 8, 2026, OpenAI began offering Ultrafast mode, which runs GPT-6.1 Sol up to eight times faster, in the API, Codex and ChatGPT Work. On October 6, it had made the Decisions API, which makes fast decisions inside apps, available in public beta to all developers.

Google Cloud Unveils Gemini agent for Work, With "Coworker" Agents That Get Their Own Email Address and a Mix of Gemini and Claude

Google Cloud Unveils Gemini agent for Work, With "Coworker" Agents That Get Their Own Email Address and a Mix of Gemini and Claude

On October 8, 2026, at its Gemini at Work 2026 event, Google Cloud announced Gemini agent, a general-purpose AI agent for work. From a single input box, you can hand it research, document creation and coding, and it keeps working on tasks that take days. It is currently in a limited preview for select companies.

Anthropic Revises Claude's Usage Policy, Banning "Persistent Abuse of AI" from November 12 and Updating Rules on Elections and Weapons

Anthropic Revises Claude's Usage Policy, Banning "Persistent Abuse of AI" from November 12 and Updating Rules on Elections and Weapons

On October 8, 2026, Anthropic announced a revision of Claude's Usage Policy. Taking effect on November 12, it bans persistent and gratuitous abuse of Claude, reorganizes the rules on elections and deceptive campaigns, and clarifies what is prohibited for weapons and surveillance.

Anthropic Releases Claude Haiku 5.5, About 75% Cheaper on Average than Haiku 4.5 with Much Stronger Computer Use

Anthropic Releases Claude Haiku 5.5, About 75% Cheaper on Average than Haiku 4.5 with Much Stronger Computer Use

On October 7, 2026, Anthropic released Claude Haiku 5.5, the small model in its Claude lineup. For requests of up to 100,000 tokens, it costs one-tenth as much as Haiku 4.5, and the amount of text it can handle has grown to 1 million tokens. It is available on Claude's web version and iOS and Android apps, and through the API.

ChatGPT Gets GPT-6 and Intelligent UI, Adding Charts, Buttons and Calculators to Answers

On October 7, 2026, OpenAI announced that it is bringing GPT-6 and a new feature called Intelligent UI to ChatGPT chats. Answers can now include diagrams, charts, buttons, forms and more that you can use right inside the conversation. The free plan will get it in a gradual rollout starting October 8.

Mistral AI Releases a Preview of Mistral Large 4, a 1-Trillion-Parameter Model, With Weights Due by Month-End

Mistral AI Releases a Preview of Mistral Large 4, a 1-Trillion-Parameter Model, With Weights Due by Month-End

On October 6, 2026, France's Mistral AI released a public preview of its new large model, Mistral Large 4. The model has 1 trillion total parameters and is now available via API, and the company says it will publish the model weights by the end of October.

Google Releases Nano Banana 2.1 Image Generation AI, With Better Text Rendering and Character Consistency and API Prices for 1K and 2K Images Cut by About Half

Google Releases Nano Banana 2.1 Image Generation AI, With Better Text Rendering and Character Consistency and API Prices for 1K and 2K Images Cut by About Half

On October 6, 2026, Google made Nano Banana 2.1, an AI model for generating and editing images, generally available. It improves image quality, adherence to instructions, character consistency across repeated edits and text rendering inside images, and is available in the Gemini app, AI Mode in Search, Flow and the API for developers.

Anthropic Expands Its Claude Startups Program With a Free Year of Claude Team and $1,000 in API Credits

Anthropic Expands Its Claude Startups Program With a Free Year of Claude Team and $1,000 in API Credits

On October 6, 2026, Anthropic announced an expansion of Claude Startups, its program for startups. Companies founded within the last five years, or that have raised funding within the last two years, can get a free year of Claude Team plus $1,000 in API credits.

OpenAI to Test Visual Ads Shown During ChatGPT Image Generation in the US

On October 5, 2026, OpenAI announced that it will test image-based ads shown while ChatGPT generates images, starting in the US in late October. The company says the ads will be clearly labeled and shown separately from the images being generated.

Gemini App Free Plan Limited to Flash-Lite from October 9 as AI Pro Gains Deep Think

Gemini App Free Plan Limited to Flash-Lite from October 9 as AI Pro Gains Deep Think

Google is reshuffling which models each Gemini app plan can use. From October 9, 2026, the free plan will offer only Flash-Lite, and AI Plus will also lose Pro (each account is notified of its switchover date), while AI Pro will gain access to Deep Think.

Anthropic Launches $100 Million Claude Frontier Academy to Train 10,000 AI Deployment Engineers

Anthropic Launches $100 Million Claude Frontier Academy to Train 10,000 AI Deployment Engineers

On October 2, 2026, Anthropic announced Claude Frontier Academy, a program to train engineers who lead AI adoption at companies. It is investing $100 million and aims to train 10,000 people by the end of 2027.

Claude Code Gets Mods, Letting Users Extend Its Behavior and Interface with TypeScript

Claude Code Gets Mods, Letting Users Extend Its Behavior and Interface with TypeScript

On October 1, 2026, Anthropic released mods, a new Claude Code feature that lets you rewrite the tool's behavior and interface with small TypeScript functions. Mods can be distributed as plugins, so teams can build extensions that fit their own operating rules.

Google Announces Gemini 4 Argon, Rolling Out First to Cyber Defenders

Google Announces Gemini 4 Argon, Rolling Out First to Cyber Defenders

On September 30, 2026, Google announced Gemini 4 Argon, its new frontier model. It will first go to a limited group of organizations working in cyber defense, then expand to paid API users and Google AI Ultra subscribers.

OpenAI DevDay 2026: GPT-6.1 Sol Arrives at One-Fifth the Price of Astra

OpenAI DevDay 2026: GPT-6.1 Sol Arrives at One-Fifth the Price of Astra

At DevDay 2026, its developer event on September 29, 2026, OpenAI announced the new GPT-6.1 Sol model, Codex Cloud, which keeps working after you close your laptop, a public beta of the Agents API, and more.

Anthropic Releases Claude Sonnet 5.5, More Than 30% Faster Than Sonnet 5

Anthropic Releases Claude Sonnet 5.5, More Than 30% Faster Than Sonnet 5

On September 28, 2026, Anthropic released Claude Sonnet 5.5, the second model in the Claude 5.5 family. The company says it generates output more than 30% faster than Sonnet 5 and cuts the cost per task by up to 30%.

Microsoft Unveils a New Copilot with Home, Code and Autopilot, Turning It into an AI You Can Hand Work Off To

Microsoft Unveils a New Copilot with Home, Code and Autopilot, Turning It into an AI You Can Hand Work Off To

On September 25, 2026, Microsoft announced a new Copilot. It rests on three pillars: Home, a starting point for your work; Code, which lets you build apps in natural language; and Autopilot, which carries out work autonomously.

ByteDance Releases the Seedream 5.0 Flash Image Generation AI, Offering the Same Editing Features as Pro Faster and Cheaper, at 0.12 Yuan per Image in China

ByteDance Releases the Seedream 5.0 Flash Image Generation AI, Offering the Same Editing Features as Pro Faster and Cheaper, at 0.12 Yuan per Image in China

In September 2026, ByteDance released Seedream 5.0 Flash, a fast, low-cost version of its image generation AI Seedream 5.0. Like the higher-end Pro, it offers partial editing by specifying locations and a feature that splits an image into up to 16 layers, and it creates images faster and more cheaply. In addition to the API for businesses and developers, it is now also available in Doubao Work, an AI for work.

Meta Connect 2026 Recap: Muse Charm, a Dedicated Muse Device, and Ray-Ban Meta Audio Glasses

Meta Connect 2026 Recap: Muse Charm, a Dedicated Muse Device, and Ray-Ban Meta Audio Glasses

At its annual Meta Connect 2026 event, Meta unveiled Muse Charm, a pocket-sized device for talking to its Muse AI agent, and Ray-Ban Meta Audio, audio glasses that pack earphone functionality and AI into one frame, among other announcements.

Google Releases Gemini 3.8 Flash TTS with More Than 2,000 Voices and Voice Cloning from 30 Seconds of Audio

Google Releases Gemini 3.8 Flash TTS with More Than 2,000 Voices and Voice Cloning from 30 Seconds of Audio

On September 23, 2026, Google released two text-to-speech (TTS) models, Gemini 3.8 Flash TTS and Gemini 3.8 Flash-Lite TTS. They offer more than 2,000 voices, let you design new voices from text instructions and can recreate a voice from 30 seconds of audio.

Tencent Releases the Hy Image 3.5 preview Image Generation AI, Handling Generation and Editing in One Model, With 2K Images at $0.024 Each

Tencent Releases the Hy Image 3.5 preview Image Generation AI, Handling Generation and Editing in One Model, With 2K Images at $0.024 Each

On September 22, 2026, Tencent's Hunyuan team released Hy Image 3.5 preview, a new AI model for generating and editing images. It creates images from text and reference images and can revise them over a multi-turn conversation. It has been built into Tencent apps such as the AI assistant Yuanbao and is also offered through the Tencent Cloud API.

Anthropic Releases Claude Opus 5.5, Cutting Typical Task Costs by About 40% Compared with Opus 5

Anthropic Releases Claude Opus 5.5, Cutting Typical Task Costs by About 40% Compared with Opus 5

On September 22, 2026, Anthropic released Claude Opus 5.5, the latest version of its top-tier model. The company says that with per-token prices cut by 20% and fewer tokens used, typical tasks cost about 40% less than with Opus 5.

Xiaomi Releases the Open-Weight AI Model MiMo-V2.6, Which Also Understands Images, Video and Audio, With Pro Topping Open Models in an External Evaluation

Xiaomi Releases the Open-Weight AI Model MiMo-V2.6, Which Also Understands Images, Video and Audio, With Pro Topping Open Models in an External Evaluation

On September 22, 2026, Xiaomi announced the MiMo-V2.6 series of AI models and released the weights for the Pro and Flash models under the MIT license. In addition to text, they understand images, video and audio together and can handle long inputs of 1 million tokens. Pro scored 46 on the overall index of the third-party evaluation site Artificial Analysis, the highest among models with published weights.

Alibaba Says Next-Generation AI Qwen 4 Is in Training, Also Unveiling Future Plans for 5 Trillion to 10 Trillion Parameters and a New AI Chip

Alibaba Says Next-Generation AI Qwen 4 Is in Training, Also Unveiling Future Plans for 5 Trillion to 10 Trillion Parameters and a New AI Chip

On September 22, 2026, at its annual technology event Apsara Conference, Alibaba announced that its next-generation AI model Qwen 4 is in training. It said the subsequent Qwen 4.5 and Qwen 5 will aim for a scale of 5 trillion to 10 trillion parameters. A release date and pricing for Qwen 4 have not been announced.

xAI Releases Grok 4.7 with Stronger Coding and Long-Running Knowledge Work

xAI Releases Grok 4.7 with Stronger Coding and Long-Running Knowledge Work

On September 21, 2026, xAI released Grok 4.7, the latest version of its flagship model. The company says it is built on a larger base model, can stay on hard tasks for longer and reviews its own work more carefully.

Alibaba Releases the Qwen-Image-2.1 Image Generation AI Model, Combining Generation and Editing in One and Creating Images With Transparent Backgrounds Directly

Alibaba Releases the Qwen-Image-2.1 Image Generation AI Model, Combining Generation and Editing in One and Creating Images With Transparent Backgrounds Directly

On September 20, 2026, Alibaba's Qwen team released the weights for Qwen-Image-2.1, a single model that both generates and edits images. The image-generating part is relatively small at 7 billion parameters, and it can create images with transparent backgrounds directly as well as edit using up to 10 reference images. The license is for research use, and commercial use requires a separate agreement.

StepFun Releases Step 5 Preview, a Flagship Model for Agents With 600 Billion Total Parameters, and Plans to Release the Weights on October 15

StepFun Releases Step 5 Preview, a Flagship Model for Agents With 600 Billion Total Parameters, and Plans to Release the Weights on October 15

On September 20, 2026, Chinese AI company StepFun (Jieyue Xingchen) announced its new flagship model Step 5 Preview and made it available via its API and other channels the same day. It is designed for uses in which AI carries out programming, specialized knowledge work and financial analysis while using tools on its own, and it can handle long inputs of 1 million tokens. The company says it will release the model weights on October 15.

Meta、AIエージェント企業Manusを最大30億ドル規模で買収 生成AIビジネス強化へ

Meta、AIエージェント企業Manusを最大30億ドル規模で買収 生成AIビジネス強化へ

米メタ(Meta Platforms)は、中国発のAIスタートアップ「Manus(マナス)」を買収すると発表しました。 買収額は公表されていませんが、関係者の話として20億〜30億ドル(約3,000億…

AI議事録サービスのNotta、AIエージェント「Notta Brain」を発表

AI議事録サービスのNotta、AIエージェント「Notta Brain」を発表

AI議事録サービスを提供するNottaは、新たにAIエージェント「Notta Brain(ノッタ・ブレイン)」を発表しました。Notta Brainは、会議内容の理解から情報検索、タスク実行までを一貫…

Google、最高峰のAIモデル「Gemini 3」を発表 – 推論能力とマルチモーダル性能で新記録を達成

Google、最高峰のAIモデル「Gemini 3」を発表 – 推論能力とマルチモーダル性能で新記録を達成

GoogleがAI技術の新たな時代を切り開く「Gemini 3」を発表しました。同社CEOのSundar Pichai氏によると、Gemini 3はこれまでで最も高度なモデルであり、複雑なアイデアを現…

AI楽曲「Walk My Walk」がビルボードチャート1位を獲得、音楽業界に衝撃

AI楽曲「Walk My Walk」がビルボードチャート1位を獲得、音楽業界に衝撃

AI生成のカントリーアーティスト「Breaking Rust」の楽曲「Walk My Walk」が、ビルボードのカントリー・デジタル・ソング・セールス・チャートで1位を獲得しました。この出来事は、音楽…

ChatGPT Atlasがリリース、AI搭載の次世代ブラウザとして登場

ChatGPT Atlasがリリース、AI搭載の次世代ブラウザとして登場

OpenAIは10月21日、AI機能を統合した新しいウェブブラウザ「ChatGPT Atlas」を発表しました。このブラウザは、ChatGPTが組み込まれた初のネイティブブラウザとして、従来の検索体験…

LovableがShopifyとのインテグレーションを発表 AIとの会話だけで本格ECサイトが数分で完成

LovableがShopifyとのインテグレーションを発表 AIとの会話だけで本格ECサイトが数分で完成

AIを活用したウェブサイト構築プラットフォームの「Lovable」が、大手ECプラットフォーム「Shopify」とのインテグレーション機能を発表しました。この新機能を使えば、プログラミングの知識がなく…

AIエージェント「Manus」がスタートアップ向けに大型支援プログラムを開始

AIエージェント「Manus」がスタートアップ向けに大型支援プログラムを開始

中国発の汎用型AIエージェント「Manus」を提供するButterfly Effect社が、スタートアップ企業向けの大規模支援プログラム「Manus for Startups」を開始しました。 プログ…

OpenAI、最新動画生成AI「Sora 2」をリリース – 音声同期とカメオ機能を搭載

OpenAI、最新動画生成AI「Sora 2」をリリース – 音声同期とカメオ機能を搭載

OpenAIは9月30日(日本時間)、次世代の動画・音声生成AIモデル「Sora 2」を発表し、米国とカナダで提供を開始した。2024年2月に発表されたオリジナルの「Sora」を大幅に進化させた本モデ…

Stability AI、エンタープライズ向け音声生成AI「Stable Audio 2.5」をリリース

Stability AI、エンタープライズ向け音声生成AI「Stable Audio 2.5」をリリース

3分間の楽曲を2秒未満で生成、ブランド独自の音響アイデンティティ構築を支援 2025年9月10日 – 画像生成AI「Stable Diffusion」で知られるStability AIは、…

「Google検索の新機能『AIモード』、日本語で正式リリース – 検索体験が大きく進化」

「Google検索の新機能『AIモード』、日本語で正式リリース – 検索体験が大きく進化」

Googleは本日、「AIモード」の日本国内での提供開始を正式に発表しました。AIモードは、最新の生成AI「Gemini 2.5」を活用した新しい検索体験であり、これまで米国や欧州などで展開されていた…

中国、AI生成コンテンツの表示義務化を9月から実施 – 世界で最も厳しいAI規制の一つ

中国、AI生成コンテンツの表示義務化を9月から実施 – 世界で最も厳しいAI規制の一つ

中国は9月1日から、AI(人工知能)が生成したすべての文章、画像、音声、動画などのコンテンツに「AI生成」であることを示す識別表示を義務化しました。規制の対象には、テキスト、画像、音声、動画、仮想シー…

AI普及で若手社員の雇用に深刻な影響 – スタンフォード大が実態調査 – 22〜25歳の若手労働者、AI関連職種で雇用13%減少

AI普及で若手社員の雇用に深刻な影響 – スタンフォード大が実態調査 – 22〜25歳の若手労働者、AI関連職種で雇用13%減少

生成AI(人工知能)の普及が労働市場に与える影響について、スタンフォード大学の研究チームが大規模な調査を実施し、衝撃的な結果を明らかにしました。ソフトウェア開発者やカスタマーサービス担当者など、AIが…

AI検索パープレキシティ、新収益分配制度で提携メディアに63億円規模還元を発表

AI検索パープレキシティ、新収益分配制度で提携メディアに63億円規模還元を発表

米国のAI検索サービス「パープレキシティ(Perplexity)」は2025年8月25日、提携するメディアや報道機関に対して新たな収益分配制度の導入を発表しました。同社のAI検索やアシスタントサービス…

Genspark、東京初の対面ミートアップを開催へ AI検索エンジンの革新者が東京でコミュニティ活性化を目指す

Genspark、東京初の対面ミートアップを開催へ AI検索エンジンの革新者が東京でコミュニティ活性化を目指す

AIエージェントを提供する米スタートアップのGensparkは、東京で初となる対面ミートアップを企画しています。このイベントは、GensparkのCOOが初来日する機会となり、東京のAIコミュニティを…

「nano banana」コードネームモデル、「Gemini 2.5 Flash Image Preview」リリース – 画像生成AIの新たな進化

「nano banana」コードネームモデル、「Gemini 2.5 Flash Image Preview」リリース – 画像生成AIの新たな進化

Googleは2025年8月26日(現地時間)、画像生成・編集AIモデル「Gemini 2.5 Flash Image Preview」(愛称「nano banana」)をGemini APIとGoo…

Google検索の「AIモード」、日本を含む180カ国に展開開始

Google検索の「AIモード」、日本を含む180カ国に展開開始

Googleは21日(米国時間)、同社のGoogle検索における「AIモード」機能を、これまでの3カ国から大幅に拡大し、日本を含む180カ国以上の国と地域で英語での利用を開始すると発表しました。 AI…

Google、生成AI「Gemini」の環境負荷を1年で大幅削減 – 1回のテキストプロンプトあたり電力33分の1・CO₂44分の1

Google、生成AI「Gemini」の環境負荷を1年で大幅削減 – 1回のテキストプロンプトあたり電力33分の1・CO₂44分の1

Googleは2025年5月、生成AIサービス「Gemini」で1回のテキストプロンプトを処理する際の環境負荷を大幅に削減したと発表しました。Google公式ブログ「Our approach to e…

アリババ、画像編集AI「Qwen-Image-Edit」を正式リリース – テキストもビジュアルも“思いどおり”に編集可能

アリババ、画像編集AI「Qwen-Image-Edit」を正式リリース – テキストもビジュアルも“思いどおり”に編集可能

生成AIの勢いが止まりません。8月19日、アリババは画像生成モデル「Qwen-Image」を大幅に強化した 画像編集特化モデル「Qwen-Image-Edit」 を正式公開しました。 テキスト修正から…

MIT調査:企業の生成AI導入、95%が失敗 – 約5兆円の投資も成果上がらず

MIT調査:企業の生成AI導入、95%が失敗 – 約5兆円の投資も成果上がらず

米マサチューセッツ工科大学(MIT)の最新調査で、企業による生成AI導入プロジェクトの95%が失敗に終わっていることが分かりました。 わずか5%のみが収益向上を実現 MITのNANDAイニシアチブが発…

謎の画像生成AIモデル「nano-banana」がLMArenaで大きな話題に – 正体不明ながら高精度で注目集める

謎の画像生成AIモデル「nano-banana」がLMArenaで大きな話題に – 正体不明ながら高精度で注目集める

AIモデル比較プラットフォームLMArenaに突如現れた画像生成AIモデル「nano-banana」が、その高い性能と謎に包まれた正体で大きな注目を集めています。 突如現れた正体不明のAIモデル na…

Google DeepMindの「Genie 3」、プロンプトからリアルタイム生成されるインタラクティブな3D世界へ一歩前進

Google DeepMindの「Genie 3」、プロンプトからリアルタイム生成されるインタラクティブな3D世界へ一歩前進

Google傘下のAI研究機関DeepMindは2025年8月5日(現地時間)、テキストプロンプトから即座に探索可能な3D世界を生成する新たなワールドモデル「Genie 3」を発表しました。研究プレビ…

OpenAI、次世代AI「GPT-5」を正式発表 – 無料版でも利用可能で「博士号レベル」の専門性を実現

OpenAI、次世代AI「GPT-5」を正式発表 – 無料版でも利用可能で「博士号レベル」の専門性を実現

米OpenAI(オープンAI)は2025年8月7日、対話型AI「ChatGPT」の次世代基盤モデル「GPT-5」を正式に発表しました。 サム・アルトマンCEOは記者会見で「対話能力が高校生・大学生レベ…

Google、大学生向けに「Gemini Pro」を1年間無料提供開始(学割)

Google、大学生向けに「Gemini Pro」を1年間無料提供開始(学割)

Googleは、大学生を対象としたAI学習支援プログラムを発表し、通常月額2,900円の「Google AI Pro プラン」を1年間無料(学割)で提供すると明らかにした。 詳細はこちら 申込み期限は…

ElevenLabs、AI音楽生成モデル『Eleven Music』をローンチ – 音楽業界との協業で権利者保護を実現

ElevenLabs、AI音楽生成モデル『Eleven Music』をローンチ – 音楽業界との協業で権利者保護を実現

AIオーディオ技術のリーディングカンパニーであるElevenLabs(本社:米国ニューヨーク州、CEO:Mati Staniszewski)は8月5日、新たなAI音楽生成サービス『Eleven Mus…

Google のAIコーディングエージェント「Jules」が正式版に – 非同期でコード修正・機能追加を自動実行

Google のAIコーディングエージェント「Jules」が正式版に – 非同期でコード修正・機能追加を自動実行

Google が開発したAIコーディングエージェント「Jules(ジュールズ)」が2025年8月6日、ベータ版を卒業し正式版として提供開始されました。Gemini 2.5 Proを搭載したJulesは…

OpenAI、5年ぶりにオープンソースAI推論モデルを発表(gpt-oss-120b、gpt-oss-20b)

OpenAI、5年ぶりにオープンソースAI推論モデルを発表(gpt-oss-120b、gpt-oss-20b)

OpenAIは8月5日(現地時間)、同社のo-seriesと同様の能力を持つ2つのオープンウェイトAI推論モデル「gpt-oss-120b」と「gpt-oss-20b」の提供開始を発表しました。これは…

Anthropic、AIモデル「Claude Opus 4.1」を発表 コーディング性能大幅向上でGPT-5との競争激化

Anthropic、AIモデル「Claude Opus 4.1」を発表 コーディング性能大幅向上でGPT-5との競争激化

米Anthropicは8月5日(現地時間)、最新のAIモデル「Claude Opus 4.1」を発表しました。同社のフラッグシップモデル「Claude Opus 4」のアップグレード版として位置づけら…

CloudflareがPerplexityの「ステルスクローラー」を告発、robots.txt無視でアクセス回避か

CloudflareがPerplexityの「ステルスクローラー」を告発、robots.txt無視でアクセス回避か

2025年8月4日、米Cloudflareは公式ブログで、AI検索サービスPerplexityがウェブサイトの「クロール禁止」指示を回避するために、未申告のクローラーと偽装したユーザーエージェントを使…

キリン、AI役員「CoreMate」を経営戦略会議に本格導入 – 12のAI人格が意思決定を支援、生産性向上と価値創造を同時実現

キリン、AI役員「CoreMate」を経営戦略会議に本格導入 – 12のAI人格が意思決定を支援、生産性向上と価値創造を同時実現

キリンホールディングスは、経営層の意思決定を支える「AI役員 CoreMate」を2025年7月以降のグループ経営戦略会議に本格導入すると発表しました。同社の「KIRIN Digital Vision…

Manusが大規模並列リサーチ機能「Wide Research」をリリース

Manusが大規模並列リサーチ機能「Wide Research」をリリース

AIエージェントプラットフォームを手がけるManusは、2025年8月1日に新機能「Wide Research」を正式リリースしました。この新機能は、複数のAIエージェントが協調・並列してタスクを実行…

Google、Gemini 2.5の新モード「Deep Think」を正式リリース

Google、Gemini 2.5の新モード「Deep Think」を正式リリース

2025年8月1日、Googleは生成AIアプリ「Gemini」に高度な推論モード 「Gemini 2.5 Deep Think」 を追加したと発表しました。まずは月額サービス Google AI U…

【速報】OpenAI「GPT-5」8月上旬リリースへ – “Mini”“Nano”を含む3ライン展開

【速報】OpenAI「GPT-5」8月上旬リリースへ – “Mini”“Nano”を含む3ライン展開

米 OpenAI は次世代大規模言語モデル「GPT-5」を 2025 年 8 月上旬に正式リリースする方針を固めました。内部試験では推論特化型の o3 モデルを組み込み、Standard・Mini・N…

Google NotebookLM、動画による要約機能「Video Overviews」を追加 – スライド付きで理解度向上を図る

Google NotebookLM、動画による要約機能「Video Overviews」を追加 – スライド付きで理解度向上を図る

米Googleは7月29日(現地時間)、パーソナライズされたAI研究アシスタント「NotebookLM」に、新機能「Video Overviews」(動画概要)を追加すると発表しました。この機能は、昨…

Genspark AIが主催する「スライド生成対決」が注目を集める

Genspark AIが主催する「スライド生成対決」が注目を集める

2025年7月29日(火)午後8時5分――AI 技術の最前線を走る Genspark AI は、グローバルイベント「GENSPARK AI 直接対決」の第2弾として「ラウンド2:スライド生成対決」を開…

Z.ai、新たなフラッグシップモデル「GLM-4.5」シリーズを発表

Z.ai、新たなフラッグシップモデル「GLM-4.5」シリーズを発表

Z.aiは2025年7月28日、推論、コーディング、エージェンシック(自律的)タスクに特化した新しいフラッグシップモデル「GLM-4.5」と「GLM-4.5-Air」を発表しました。 2つのモデルライ…

Sales Marker、国内初のAIスーパーエージェント「Orcha」をリリース – 複数AIツールの統合課題を解決

Sales Marker、国内初のAIスーパーエージェント「Orcha」をリリース – 複数AIツールの統合課題を解決

営業・マーケティング自動化を手がける株式会社Sales Markerは、AIスーパーエージェント「Orcha(オルカ)」を正式リリースしました。同社は「国内初」としており、第一弾として自動スライド生成…

NVIDIA研究チーム、「小型言語モデル(SLM)がエージェント型AIの未来」と提言

NVIDIA研究チーム、「小型言語モデル(SLM)がエージェント型AIの未来」と提言

NVIDIA Researchの研究チームが、AIエージェントシステムにおいて小型言語モデル(SLM:Small Language Models)が大規模言語モデル(LLM:Large Languag…