AI2 unveiled Asta, a suite of agentic tools for scientific research, including:

—Asta agents to assist researchers with scientific tasks
—AstaBench suite & leaderboards for evaluating agents
—Asta resources software components to create and extend agents

Anthropic just launched Claude for Chrome, enabling agentic browsing via a new extension

It is being piloted via a waitlist exclusively for 1,000 Claude Max subscribers in a limited preview

Comet and Dia are soon going to have some serious competition

Now you can forward your emails to Manus AI so it can deal with them instead of you.

Cold emails are cooked 👀

Krea AI is opening a waitlist for a new Real-time Video generation model.

"Behind the scenes, our approach is motivated by modern “world model” ideas: systems that learn how scenes evolve and how actions ripple forward in time."

Stitch by Google got a Canvas feature! Now you can plot and edit a whole UI journey over there.

So much to test 🤯

* Attaching a proper video

Ex-Musk engineer stole Grok code and joined OpenAI

Before leaving, he sold $7M worth of xAI shares. Elon Musk is now suing him for allegedly stealing Grok’s source code.

Vote.

OpenAI just published their official prompting guide for GPT-5

Source - https://cdn.openai.com/API/docs/gpt-5-for-coding-cheatsheet.pdf

#prompt

Microsoft just unveiled its first in-house AI models: MAI-Voice-1 and MAI-1-preview

MAI-Voice-1 generates a minute of speech in a second

Lol it generates minutes of voice in seconds😂

OpenAI launched gpt-realtime speech-to-speech model with remote MCPs and image support

It has nuanced abilities like detecting nonverbal cues and switching languages while keeping a natural conversation

Scored 82.8% on audio reasoning benchmarks

xAI introduced Grok Code Fast 1, a fast reasoning model for agentic coding tasks

It uses a lightweight architecture for cost efficiency and is strong in TypeScript, Python, Java, Rust, C++, and Go

Free on agentic coding platforms for a limited time

Higgsfield AI launched Higgsfield Speak 2.0, a new upgrade to its custom avatar tool

It brings more realistic motion, advanced lip-sync, and enhanced video control

Also supports 70+ languages

Apple launched two new open models: FastVLM and MobileCLIP2

The models are up to 85x faster and 3.4x smaller than previous work

Suitable for high-res image processing tasks like OCR, image captioning, visual question answering, and image understanding

Microsoft just launched a larger 10B parameter version of its VibeVoice text-to-speech model

Available under MIT license, it generates multi-speaker podcasts (going up to 45 minutes) in minutes

Supports up to 32K context

Tencent launched two new Hunyuan models:

—HunyuanWorld-Voyager, an ultra-long-range world model with native 3D reconstruction and memory

—Hunyuan-MT-7B and Hunyuan-MT-Chimera-7B, a joint AI translation system that outperforms rivals across 33 languages