Qwen has just released a model on par with GPT-4o...

And you can run it locally easily 🤯

Yep. GPT-4o level AI running offline on a laptop.

- Fully open source
- Only 3B active parameters
- 262k context length natively

Quick steps to run it on your machine and details below 1. Download LM Studio and the model

- Install LM Studio for your OS (macOS, Windows, Linux)
- In the search tab type "Qwen3 30B A3B 2507"

I highly recommend the versions quantized by Unsloth, especially those marked "UD".

They're REALLY efficient. E.g.: Q4_K_XL Qwen3 30B A3B Instruct 2507 UD

Then just click on the download button.

Wait so Alibaba Qwen has just released ANOTHER model??

Qwen3-Coder is simply one of the best coding model we've ever seen.

→ Still 100% open source
→ Up to 1M context window 🔥
→ 35B active parameters
→ Same performance as Sonnet 4

They're releasing a CLI tool as well ↓ You can use the model for free on Qwen Chat:

1. Create a free account
2. Select Qwen3-Coder in the list
3. You can also upload files (codebase)

Wait NVIDIA has just released new SOTA open source models?!

Available in 4 sizes 1.5B, 7B, 14B and 32B that you can run 100% locally.

- OpenReasoning-Nemotron
- SOTA scores across many benchmarks
- Tailored for math, science, code

How to run it on your laptop and details below You can run them using LM Studio for free:

1. Download LM Studio for macOS, Windows or Linux
2. In the search tab, type "openreasoning"
3. Install the version you want

I suggest the 7B (very good for its size) from Bartowski in the Q4_0 quantization if you're using an ARM processor like me.

Grok 4 is excellent for vibe-coding

You can give it your entire codebase in 5s to add features or debug.

→ No need for Cursor
→ Works directly on the Grok site

Quick steps below: 1. Generate command

To export your codebase Grok 4 can generate the command that you can reuse over and over again.

Prompt template:

"Give me a shell command I can use in my [IDE Name], with my [Project Type, eg. Node js] to download the entire codebase in a single text file but without the node_modules, package-lock.json, or hidden files/folders, just the really useful ones."

Gemini CLI can automate your computer using MCP 🔥

Add Windows MCP (or macOS MCP) to Gemini CLI and you can tell it what to do autonomously.

Gemini then takes control of your entire system to achieve the goal you've set.

Links below As a reminder, you have 1'000 free requests PER DAY using Gemini CLI!

Windows MCP used → github.com/CursorTouch/Wi…

MacOS MCP (never tried) → github.com/baryhuang/mcp-…

You can add a local "=AI" formula in Excel

Excel can then process data using a free and open-source AI model like Gemma.

This means that it understands what's in the cells and returns a tailored response based on your prompt even offline.

Steps to set it up and examples below Keep in mind that everything runs locally, even the AI model (LLM): no remote servers or APIs!

EXAMPLE 1 - Categorize data

Formula:
=AI("Is this a basketball or baseball team? Just write one word: Basketball or Baseball. Team name: ",A1)

Finally! Google has just released Gemini CLI an AI agent that brings Gemini directly into your terminal

→ 1,000 free requests PER DAY
→ Open source

You can use it as a coding agent, automate tasks, use MCPs, generate videos & images, etc.

Steps to install and use it: 1. Install Node

- Go to
- Install the version for your OS

The easiest way is to download a prebuilt version for Windows or macOS (see video). nodejs.org/en/download

Google Jules is the most underrated vibe coding tool

You can start a project with Replit and then assign tasks to Jules to perform autonomously 🔥

1. Replit Agent create the project
2. Create a GitHub repo for your "code"
3. Link Jules (free) to this repo, done!

Steps below: 1. Link Replit and GitHub

After creating your project with Replit Agent:

- Click on the icon in the left-hand column
- Select "Git"
- Create a new repo or link an existing one

Note: Replit offers to create a repo if you don't have one.

All you need is a free GitHub account.

Google has released an open source project for building fullstack agents with Gemini 2.5 🔥

You can run and modify it with Replit to create your own agent.

By default the agent has:

- A full conversational interface
- A backend for an AI-powered research

Steps to use it below: 1. Import in Replit

- Go to Replit.com
- Select the import option then GitHub
- Click on import and paste the link of the repo:

github.com/google-gemini/…

DeepSeek has just released a powerful new 8B model 🔥

This means you can run it locally and for free on a laptop with zero limitations.

You just need LM Studio and 4Gb of RAM.

The few quick steps below: 1. Download LM Studio (free)

Go to and download the version for your OS.

Available for:

- Windows (x86/ARM)
- macOS (M series)
- Linux lmstudio.ai

This AI agent is a powerful data analyst

It can search, process, complete and visualize your data in a single chat.

You can import files or start from scratch.

Step-by-step process with Genspark: 1. Import or search data

Go to and select “AI Sheets”.

You can either :

- Import your files
- Search for data

In the 2nd case, the agent aggregates and formats the data according to your criteria. genspark.ai

Google has quietly released AI tools for language learning

You can access these 3 tools for free:

1. Find vocabulary for any situation
2. Learn idioms from a generated conversation
3. Talk to the camera about your surroundings

Details and access: 1. Tiny Lesson

Choose your language and a situation.

Gemini will then generate customized vocabulary, phrases and tips for you to listen to and learn.

AI Agents are becoming extremely capable

Among other things, in a single prompt you can:

- Search for information
- Compile insights
- Create visuals
- Generate slides

Genspark’s Super Agent manages all the tools by itself.

(More below) You can use several agents for a single request.

Go to the "All Agents" section of :

→ Deep Research
→ AI Slides
→ Call For Me (yes!)
→ Agentic Fact Check
→ Agentic Data Table

I'll show you a few combos possible using @genspark_ai ↓ genspark.ai

Anthropic has released a full guide on agentic coding

There are 6 points to maximize the efficiency of Claude Code and LLMs to build.

Everything you need to know below

(And link to the official blog post) 1. Customization & optimization

- Create CLAUDE .md files to document workflows, commands, and coding conventions.

- Curate a list of allowed tools for more efficient automation.

- Install the GitHub CLI for better GitHub integration.

ChatGPT o3 and canvas is a builder's dream

You can create a complete web app with a backend and deploy it.

The whole workflow takes less than 5 minutes, here are the steps: 1. Prepare the prompt

I use my GPT "Prompt Enhancer" to save time.

It transformed my basic prompt into a much more complete prompt so that o3 could one-shot the app.

(You can get it for free below)

Prompt:

"You're a senior front-end developer with experience in building productivity tools using vanilla JavaScript and browser-based storage. Create a single-file web application that serves as a "Prompt Organizer" to help users save, edit, delete, and categorize AI prompts directly in the browser without any external dependencies. Use only HTML, CSS, and JavaScript, and store data persistently using localStorage or IndexedDB to ensure saved prompts remain available across sessions. Each prompt should include fields such as title, description, tags, and the full prompt text. Users should be able to filter prompts by tags, search by keywords, and sort by date created or modified. Include example prompts to populate the organizer on first load. Structure the code so it’s readable and modular within the single HTML file, using internal