Grok 3 is Here - Is It Any Good?🧐

You're receiving this email because you registered for one of our workshops. You can unsubscribe at the bottom of each email at any time.

Hey there!

Remember when I told you the integrations block were just the beginning of MindStudio entering the agentic era? Well, I don’t make empty promises. We released over a dozen integration blocks and launched a major interface update: you can now run MindStudio workers from wherever you want using our Chrome Extension.

And not only that. We’re gonna pay you to build good agents for the Chrome Extension. Register to our kickoff webinar to learn more :)

In other news, xAI launched Grok 3, NVIDIA is back to a $3.4t valuation after a sharp decline due to DeepSeek R1 (now available hosted in US data centers), the EU commits $200b to AI investment in Paris, and OpenAI hints at an open source o3-mini.

Continue reading to learn more!

As a reminder, we’re now welcoming partners who want to build AIs for their clients. Sign up for extra support, training resources, and more here.

🗞️ Industry news

xAI releases Grok 3. Is it the new AI leader?

This week, xAI released Grok 3, the leader of the chatbot arena and its latest SOTA model. At first, the model was only available for X Premium+ subscribers, but is now free for everyone to test. The rate limit seems to be 25 messages per hour.

The xAI team is hyping it up as the best model available using a few benchmarks as proof. Some believe it, others don’t.

Grok 3 comes with reasoning and deep research, features copied from its big brothers ChatGPT and Perplexity. Grok is the only model that can do proper deep research on X threads, making it somewhat unique.

Here’s what we know so far:

• Regardless of what you think of Musk, now one of the most controversial people on the planet, the xAI team is very talented and managed to release a truly great model. After testing it, it’s clear the model is very capable and likely matches or surpasses the “vibe” of GPT-4o for generic asks

• Deep Research in Grok 3 is nowhere near as good as OpenAI’s deep research, but it does provide good results and rivals Perplexity’s Deep Research. Perplexity is much cheaper, so they will both coexist at different price points

• X increased the cost of Premium+ significantly, now hovering at the $40-60 price point. The price depends on your region. That’s more expensive than any other commercial chatbot solution other than ChatGPT Pro (at $200/m)

• Grok 3 is an uncensored SOTA model. This never really happened before, and it has the potential to create issues other models can’t. It will refuse very harmful requests, but the bar here seems obscenely low. It will do pretty much whatever you ask it to do unless it has the potential for real-world danger. Whether this is a pro or a con is up to you

• Deep Research in Grok is awesome if you work in AI, Web3, or tech-first niches. It’s the only real way to get data from X easily and effectively

• The “Thinking” mode exposes the Chain of Thought. Musk mentioned it’s not the full CoT, others in the team said it is. Either way, it’s more transparent than what most others provide

So… is Grok 3 my new default AI?

No, not really. At least not for now.

While it does work well, and has unique features I enjoy and use from time to time, it’s still not the bump needed to switch to a new platform. I’d definitely use it via API, though, as soon as it’s available. It’s the best model for X searching tasks.

The EU commits to spend $200b in AI. Will they?

The EU tried to ride the wave of Stargate in the US to launch an equally impressive €200b investment project for the block, claiming Generative AI itself can add $575b to the European economy by 2030.

The initiative gathered support from all major EU tech firms such as Adyen (Stripe alternative), SAP (dinosaur in the service space), ASML (leader in chip production), Mistral (the only important AI lab in EU), and many more.

Over 70 signed the pledge, which is a mix of public and private investment goals, with a total market cap of $3t. It’s a multi-year commitment to AI growth, starting with AI gigafactories.

Now, a few issues:

• Just like Stargate, this money doesn’t exist right now. It’s merely a promise or a faint goal

• The EU doesn’t have a money problem, it has a management problem. The reason all major tech companies are Americans is not because Europeans aren’t ambitious or because they’re uneducated. There are more developers in the EU than in the US, for example. It’s a systemic, top-down issue, one that EU politicians have failed to address for decades at this point

• Mario Draghi, Italy’s ex-prime minister and ex-CEO of the central bank, already said the EU needs investments up to $800b or more per year to catch up with US’ growth. And, without growth, the social model cannot last for long. Western EU in particular is plagued by laughably low growth rates with massive brain drain

• For my American friends, the “EU commission” is not the same as the US government. They have some power, but they mostly use it to regulate, not to advance agendas. Even investments such as this need to be managed by multiple countries agreeing with each other on initiatives. It can’t, and won’t be, a cohesive initiative.

Let’s see how this develops, and which (if any) of the big investments promised by the US and EU becomes a reality.

What’s going on with Chinese models?

DeepSeek R1 (partially open source) and Qwen 2.5 Max are both in the top 10 best models according to chatbot arena. We added DeepSeek R1 to MindStudio as well due to the incredible request, and the company tanked NVIDIA’s stock by over 15%.

Here are my two cents:

• DeepSeek R1 is an insanely good model for the price. While o3-mini is also a great deal, R1 is still raw open source intelligence you can run at multiple scales, including locally, for very cheap. I use it in Perplexity (hosted in the US) and MindStudio

• Qwen 2.5 Max is also very impressive. Coming from Alibaba, it definitely has a bigger scope in mind and is near the ceiling of SOTA. Nothing that stands out enough for me to consider using it long-term, though

• The whole DeepSeek story was mostly hype from journalists that don’t know much about the “cost” of AI. The $6m figure thrown around was simply for training, for example, not the actual cost to go from 0 to R1. Additionally, the narrative that DeepSeek is somehow a public benefit company is equally funny, given the owner of DeepSeek is an hedge fund in China

• Expect these models to continue being biased towards the CCP’s agenda. After all, any server in China needs authorization from the government, and they have much stricter censorship rules.

So, yes, Chinese companies are making great strides in AI. Experts like Dario Amodei (CEO @ Anthropic) believe US companies shouldn’t export chips to limit the progress.

🔥 Product Updates

MindStudio agents can now run in your Chrome browser - and they’re awesome! We held two intro webinars about it, and if you missed them, the replays are on YouTube.

Our new Chrome Extension lets you run MindStudio agents directly in your browser, using the context of the web page you're on. This feature opens up a world of possibilities for real-time, context-aware AI interactions.

• Pre-loaded Agents: we've added several pre-loaded agents that you can use without any cloning required—just click a button and start using them. Among them is the "Researcher" agent, which functions similarly to OpenAI's Deep Research. It can suggest topics from the page you're on or take your input to perform deep research and provide comprehensive outputs

• New Integration Blocks: we've expanded our integration capabilities with new blocks for YouTube and email enrichment. You can now scrape captions from YouTube videos, retrieve YouTube data, and find emails using first and last names with company details via hunter.io

• Agent Submission Program: we're excited to announce that we'll be paying users for submitting high-quality agents to our store. This isn't a hackathon—take your time to build something unique and valuable. More details will be shared in our upcoming kickoff. Sign up here

We also released major updates to the UI to make it work perfectly in the sidepanel. Now, you can do pretty much anything you could do before but from the Chrome extension.

The extension is a fundamentally different way to think about AI. Now, instead of the standard Chat interface or a custom coded one (which required coding experience), you can build beautiful experiences that run where and when you need them to.

The AI models paired with loops, logic, dynamic variables, web search, dozens of integrations, and a block to generate HTML on the go to display the result beautifully are a new paradigm for us and for the AI space as a whole.

We’re sure you’re going to find awesome ideas to build!

💡 Tip of The Week

There are a few notable models that we don’t see used very often which definitely deserve your attention:

• Gemini 2.0 Flash & Flash Lite are awesome models for very affordable rates. They’re my new default for workflows that don’t require high intelligence. They’re fast (up to 150tks/s), cheap, and effective

• The new Sonar Pro models come with citations and are great for quick web searches and grounding

• For intense reasoning tasks, give a go to o3-mini. It’s not basically cheaper than GPT-4o but is faster and smarter.

🌯 That’s a wrap!

Stay tuned to learn more about what’s next and get tips & tricks for your MindStudio build.

You saw it here first,

Giorgio Barilla
MindStudio Developer & Project Manager @ MindStudio