🚀 New open-source AI beats GPT-5.2

Good Morning! Today, you’ll learn how to setup and use Claud Skills to automate repetitive tasks, build reusable workflows, and connect Claude to real apps like Gmail.

Plus, in today’s AI newsletter:

• Moonshot AI Releases Kimi K2.5 (Open-Source)

• Google Brings Gemini Auto-Browse to Chrome

• How to Setup And Use Custom Claude Skills

• 4 new AI tools worth trying

AI TOOLS

🧭 Google Brings Gemini Auto-Browse to Chrome

Google is rolling out an “auto browse” feature in Chrome that allows Gemini to act as an AI agent, completing complex tasks like shopping, research, and scheduling on your behalf. It’s launching for AI Pro and Ultra subscribers in the US.

• Can research flights, hotels, appointments, and complete forms end-to-end

• Shops within a budget: finds items, applies coupons, and adds to cart automatically

• Uses Chrome’s password manager to log in when required

• Deep integrations with Gmail, Calendar, Maps, Shopping, and Flights

This pushes Chrome from a passive browser to an active AI agent. If Gemini reliably executes real tasks, not just suggestions, Google could define how people actually use the web in an agent-first era.

AI MODELS

🌕 Moonshot AI Releases Kimi K2.5 (Open-Source)

Kimi K2.5 is Moonshot AI’s latest open LLM, built on a mixture-of-experts design and trained on 15 trillion tokens, including multimodal data. The company says it’s optimized for faster reasoning, agent-based workflows, and lower hardware usage.

• 1T-parameter MoE model with only ~32B active parameters per task

• Trained on 15T tokens (text + multimodal), improving chart and file understanding

• “Thinking mode” + Agent Swarm can orchestrate up to 100 agents per prompt

• Outperformed GPT-5.2 on HLE-Full and matched top models across 20+ benchmarks

This is another strong signal that open models are closing the gap with frontier systems. With agent swarms, parallel attention, and top-tier benchmark results, Kimi K2.5 positions China’s open AI ecosystem as a serious contender in advanced reasoning and agentic AI.

HOW TO AI

💻 How to Setup And Use Custom Claude Skills

In this tutorial, you’ll learn how to setup and use Claud Skills to automate repetitive tasks, build reusable workflows, and connect Claude to real apps like Gmail.

🧰 Who is This For

• Professionals who want to save time on repetitive tasks

• Creators who want reusable AI workflows instead of rewriting prompts

• Developers and power users exploring agent-based AI

• Anyone using Claude Pro or Team and wanting to unlock advanced automation

STEP 1: Turn On the Skill Creator

Before you can build skills, you need to enable Claude’s advanced creation features.

Start by clicking your Profile icon in the bottom-left corner and opening Settings. Inside the settings menu, find the Capabilities tab and open it.

Scroll down and make sure the following switches are turned ON (green):

• Skill Creator (this is the most important one)

• Theme Factory

• Artifacts Builder & Canvas Design

If you see Skill Creator, double-check that it’s enabled. Without this, none of the next steps will work.

STEP 2: Create Your First Custom Skill

Now it’s time to build a real skill. As an example, we’ll create an AI FOMO Summarizer that helps you decide whether a new AI tool is worth your time.

Start a new chat by clicking the Claude logo or the + button.

In the message box, type, “Make a new skill” and press Enter.

Claude will switch into an interview mode and ask you a few questions. Answer them like this:

• What should this skill do?
  → “This is an AI FOMO Summarizer. Using an AI feature announcement or press release, it will perform its own research and present a score of how valuable it would be for me to spend time learning and using it based on what Claude knows about me.”

• When should I use it?
  → “An example is OpenAI’s new web browser Atlas or Claude’s new Agent Skills announcement. I’d say “ai fomo - OpenAI Atlas https://openai.com/index/introducing-chatgpt-atlas/”

• What makes it unique?
  → “Use what you know about me from our conversation history to determine relevance. But if you don’t have enough insight from that, ask me up to 3 questions for clarity.”

Before Claude finalizes the skill, add one important follow-up question:
“What else should I clarify?”

This step helps Claude fill in any missing logic and makes the skill much more accurate.

STEP 3: Install the Skill (Download & Upload)

Once Claude finishes, it will generate a skill file for you directly in the chat.

Click the Download button on that file.

Next, go back to Profile → Settings → Capabilities.

Scroll down to the Skills section and click Upload Skill. Drag and drop the file you just downloaded into the upload box.

If everything worked, you’ll see the new skill appear in your list. From now on, you can use it in any new chat just by pasting a link to an AI tool.

STEP 4: Connect Real Apps (Like Gmail)

Skills become even more powerful when they can access real data.

For example, let’s try it with Gmail.

To connect Gmail, open a new chat and click the Tools icon inside the message box (it looks like a wrench or search bar). Choose Gmail Search and sign in to your Google account.

Once connected, type, “Make a new skill”, again.

This time, give Claude instructions like:
“Check my unread emails from yesterday and draft replies to the important ones.”

After Claude creates the skill, repeat the same install process:
Download the file → go to Settings → Capabilities → Upload Skill → upload it.

Now every morning, you can simply type something like “Gmail Summary”, and Claude will automatically read your actual inbox and help you respond, no repeated prompting required.

Security researchers warn Moltbot, previously Clawdbot, requires a specialist skillset to use safely, as data exposure risks persist even when set up correctly.

US-based AI startup Arcee releases Trinity Large, a 400B-parameter open-weight model that it says compares to Meta's Llama 4 Maverick 400B on some benchmarks.

OpenAI is considering using biometric verification like World's eyeball scanning Orb for its planned social network to ensure its users are real people.

OpenAI is in talks to raise up to $30B from Nvidia, less than $10B from Microsoft, and $10B-$20B from Amazon as part of a $100B funding round.

🦞 Moltbot: An open-source AI assistant that works inside chat apps

🚀 Kimi K2.5: Moonshot AI’s powerful open-source model built for agents

📄 Prism: OpenAI’s all-in-one research workspace

🗣️ Qwen3-TTS: Alibaba’s open-source text-to-speech models

THAT’S IT FOR TODAY

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