TestingCatalog AI News 🗞 – Telegram
TestingCatalog AI News 🗞
4.42K subscribers
2.81K photos
355 videos
40 files
3.82K links
Reporting AI nonsense. A future news media, driven by virtual assistants 🤖
Download Telegram
xAI introduces Grok 4 and Grok 4 Heavy as its most powerful AI models yet

xAI has released Grok 4 and Grok 4 Heavy via premium subnoscriptions, offering advanced reasoning, multimodal input support, and strong benchmarks. Built on the Colossus supercomputer, Grok 4 targets technical users with coding and science capabilities.

🗞 #grok
🔥6👏1
Countdown starts for Deep Think rollout while Agent Mode surfaces in code

Google is testing Deep Think, a new AI model showing improved output but with slower response times. Features like Bespoke, Learning Mode, and Agent Mode also suggest a broader focus on personalized and autonomous AI tools.

🗞 #gemini
🔥9👀1
Microsoft develops Copilot 3D experiment targeting 3D creators

Microsoft is testing new Copilot features like character avatars and 3D model generation, but most remain in early stages. UI tweaks and previews show progress, though key functions such as image-to-3D remain unreliable and limited to test environments.

🗞 #microsoftcopilot
3
NotebookLM in testing for interactive Quizzes and direct Word support

Google is preparing updates for NotebookLM, including AI-generated quizzes, video overviews, and support for Word uploads. These features target varied learning styles and study needs while improving compatibility and sharing capabilities.

🗞 #notebooklm
👍61
Grok debuts interactive AI Companions on iOS with anime avatars

Grok introduces AI Companions to its iOS app—3D characters with voice chat and level-based features. With avatars like Ani and Rudy, users can unlock new character modes, signaling Grok’s push into gamified, character-driven AI experiences.

🗞 #grok
🔥7🗿1
BREAKING 🚨: According to NYT, a new Superintelligence Lab is considering to drop Behemoth in favour to developing a new closed source model.

Meta is on a streak 👀

* Want to try a new “news” format
👍13
BREAKING 🚨: Thinking Machines has raised $2 billion with participation from NVIDIA and AMD and about to present its first product soon.

What would it be? 👀
🤔9👍1
Mistral released Voxtral: open-source audio models for businesses

Mistral launched Voxtral, its first open-source audio model family, offering trannoscription and speech comprehension with two scalable variants. It supports multilingual input, local deployment, and API use at competitive pricing for enterprise use.

🗞 #mistral
🔥5🥰1👏1
Mistral AI to add Dictation, Memory, Projects and Research tools to Le Chat

Mistral AI is preparing key updates to its Le Chat platform, including voice input via Voxtral, a research planning tool, memory control, and project organization features—moves aimed at rivaling leading AI platforms like ChatGPT and Gemini.

🗞 #mistral
6🔥2👍1
OpenAI readies Agent Mode for ChatGPT to challenge Microsoft Office

OpenAI is developing an Agent Mode for ChatGPT that combines app integration, browser control, and workflow automation to position it as a central productivity tool, directly challenging Microsoft Office's role in enterprise environments.

🗞 #chatgpt
👍61
Google tests Guided Learning in Gemini as AI education race heats up

Google and OpenAI are testing educational tools—Gemini's Guided Learning and ChatGPT's Study Together—aimed at structured knowledge support, though both remain inactive and lack clear differentiation in current testing phases.

🗞 #gemini
🔥7❤‍🔥11
Le Chat update brings Deep Research, multilingual reasoning, and Projects

Mistral AI's Le Chat now offers Deep Research for fast, reference-backed briefs, voice input via Voxtral models, multilingual reasoning with Magistral, and project folders to maintain context, all available free on web and mobile.

🗞 #mistral
5🔥2
OpenAI unveils Agent Mode in ChatGPT for complex task handling

OpenAI launched ChatGPT Agent on July 17, 2025, combining browsing, coding, and file editing within a virtual computer. It performs end-to-end tasks from a single prompt, with consent-based controls and strong performance in complex tasks.

🗞 #chatgpt
🎉5🔥2👏2
Anthropic develops interactive Study Projects to turn Claude into a tutor

Anthropic is developing a Study Projects feature for Claude focused on students, offering structured learning paths and adaptive support. This move aligns with trends across major AI firms aiming to address educational use cases amid growing scrutiny.

🗞 #claude
👍8
Signs of GPT-5 emerge in Biosec repo as OpenAI nears release

Early evidence suggests GPT-5 is undergoing internal testing, appearing in biosecurity research and outperforming in coding benchmarks. OpenAI signals a cautious rollout amid scrutiny and rising expectations for advanced reasoning capabilities.

🗞 #chatgpt
🔥7🥰1🆒1
Grok may soon get model slider with Expert, Heavy and Fast options

xAI is testing a web-based model selector for Grok with Fast, Expert, and Heavy options, hinting at plans to support broader user needs through flexible performance tiers and subnoscription-based access, including a possible desktop version.

🗞 #grok
🔥7👍1
Media is too big
VIEW IN TELEGRAM
BREAKING 🚨: A huge cluster of Comet invites will be approaching the Earth soon!

Will be dropping some of the Comet invites here randomly. You are not prepared! ☄️
16🔥6🍾5