Happy Sunday! We just had another crazy week in AI. Kimi AI has released the biggest and most capable open-source model yet, while xAI has launched Grok automations.
And that's not all, here are the most important AI moves you need to know this week.
NotebookLM is gone. Google renamed its research tool to Gemini Notebook and paired the rebrand with its biggest upgrade yet: every notebook now gets a secure cloud computer that writes and runs code against your sources. Upload your docs, ask a question, and instead of guessing at numbers from text, it actually computes the answer.
Real Data Analysis: The old NotebookLM could only read your sources and answer in words. The cloud computer writes and executes code natively, so it can clean spreadsheets, crunch numbers, and test whether a pattern actually holds across your uploaded files.
Deeper Google Integration: Notebooks now sync with the Gemini app, and Google plans to embed them directly into AI Mode in Search, so your research follows you across the whole ecosystem.
Rolling Out Now: The cloud computer is live for Google AI Ultra subscribers and eligible Workspace business customers, with all AI Pro subscribers on the web getting access over the coming weeks.
Try it now → https://notebooklm.google.com
Reve just launched Reframe, a tool that takes any image and pushes the frame outward to reveal more of the scene or zooms in tighter on what matters. You pick the aspect ratio, and the model fills in the rest. No cropping, no regenerating from scratch, no losing the shot you already liked.
Two Directions: Expand outward to build out the world beyond the edges of your image, or punch in closer to isolate a detail, all from a single source image.
Any Aspect Ratio, Zero Cropping: Turn a 1:1 product shot into a 9:16 Story or a widescreen banner without cutting out what matters. Your original composition stays intact while the frame changes around it.
Built Into the Reve Workflow: Reframe sits alongside tools like Relayout and Color Control, so you can generate an image once and adapt it for every placement instead of re-prompting from zero.
Try it now → https://reve.com
xAI has shipped Automations for Grok. Describe any recurring job in plain English, set a schedule or a trigger like an incoming email, and Grok executes it automatically from then on, whether that's research finished before you wake up or an important email flagged the moment it lands. Scheduled automations are free for everyone.
Plain English Setup: Instructions read like a normal chat message. Attach files for context, add connectors and skills so one automation can pull data from your tools and post results to places like Slack, then set it to run once, daily, weekdays, weekly, monthly, or yearly in your timezone.
Email Triggers: SuperGrok subscribers can point an automation at their inbox and have it fire whenever a matching email arrives, filtered by sender, recipient, or subject line.
Full Run History: Every execution starts a fresh conversation with current data, saves the complete thread to your run history, and reports back by email, app notification, or both.
Try it now → https://grok.com/automations
Thinking Machines Lab, the startup founded by former OpenAI CTO Mira Murati, released Inkling: a 975 billion parameter Mixture-of-Experts model with full weights on Hugging Face. The pitch isn't "strongest model on Earth." It's a foundation you download, run yourself, and fine-tune on your own expertise so it beats generic models at your specific tasks.
Serious Specs: 975B total parameters with only 41B active per token, natively multimodal across text, images, and audio, trained on 45 trillion tokens, with a context window of up to 1 million tokens.
Actually Open: Apache 2.0 license means you can download, modify, self-host, and commercialize it royalty-free, with fine-tuning available today through the company's Tinker platform.
Best US Open Model: Inkling debuted at 41 on the Artificial Analysis Intelligence Index, the highest score ever for a US open-weights release, ahead of Nemotron 3 Ultra (38) and gpt-oss-120b (24).
Try it now → https://thinkingmachines.ai/inkling/
PrismML, a Caltech spinout backed by Khosla Ventures, Google, and Samsung, just compressed a 27 billion parameter model down to 3.9 gigabytes. It runs locally on an iPhone 17 Pro at roughly 11 tokens per second, keeps around 90% of the full model's benchmark performance, and the whole thing is free under Apache 2.0. No cloud, no latency, no data leaving your device.
Extreme Compression: The same model in standard 16-bit precision needs about 54 GB. PrismML's 1-bit build shrinks it to 3.9 GB for phones, while a 5.9 GB ternary variant targets laptops and retains roughly 95% of full-precision performance.
Full Capability On Device: It's multimodal with vision support, handles tool calling and agentic loops, and keeps the base model's 262K token context window, all running privately on your own hardware.
Free and Open: Both variants ship under Apache 2.0 with weights on Hugging Face, support for llama.cpp and MLX, and a free developer preview API if you want to test before installing anything.
Try it now → https://prismml.com
Moonshot AI released Kimi K3: a 2.8 trillion parameter Mixture-of-Experts model with a 1 million token context window and native vision, making it the largest open-weight model ever built. Within hours of launch it jumped from #18 to #1 on Arena.ai's Frontend Code Arena, ahead of every leading closed model from the US.
Frontend Coding Champion: K3 tops the Frontend Code Arena with a score of 1,679, beating Claude Fable 5 (1,631) and GPT-5.6 Sol (1,618) in blind human preference testing, and ranks first in six of the seven frontend domains measured.
Massive but Efficient: The 2.8T parameter model activates just 16 of its 896 experts per token, about 1.8% of the pool, and uses 21% fewer output tokens than its predecessor K2.6 on equivalent tasks.
Open Weights Incoming: K3 is live now on kimi.com and via API at $3 per million input tokens and $15 per million output, with the full weights publishing under a modified MIT license by July 27.
Try it now → https://www.kimi.com
Thanks for making it to the end! I put my heart into every email I send. I hope you are enjoying it. Let me know your thoughts so I can make the next one even better.
See you tomorrow :)
Dr. Alvaro Cintas
✓ Full archive of premium guides with ready-to-use prompts
✓ Structured AI courses (step-by-step, start-to-finish)
✓ Every upcoming premium tutorial