Mistral Drops New AI Models for Laptops and Phones "Les Ministraux" - The AI Podcast

Summary7 min read

The AI Podcast: Episode Summary

Episode Title: Mistral Drops New AI Models for Laptops and Phones "Les Ministraux"
Release Date: March 25, 2025
Host: The AI Podcast

Introduction to Mistral AI's Latest Release

In this episode, The AI Podcast delves deep into the groundbreaking advancements made by Mistral AI, a Paris-based startup that is rapidly becoming a significant player in the artificial intelligence landscape. The focus of the discussion centers on Mistral’s newly released AI models tailored for edge devices—specifically laptops and smartphones. These models, collectively named LE Ministro, mark a pivotal shift towards more accessible and efficient AI applications.

[00:02] Host: "We have big news from Mistral AI as they have just released their first new model that's specifically designed to run on laptops and phones or what we're calling Edge cases."

Overview of LE Ministro Models

Mistral introduced two primary models under the LE Ministro family:

Ministral 3B
Ministral 8B

Both models boast an impressive 128,000-token context window, enabling them to process inputs equivalent to a 50-page book. This vast capacity ensures versatility for a wide range of applications, catering to the needs of the majority of users.

[00:10] Host: "Both of them have a context window of 128,000 tokens. So this means that they can both take in about the length of a 50 page book."

Hybrid Model Approach for Enhanced Efficiency

One of the standout features of LE Ministro is Mistral’s innovative hybrid model strategy. By integrating smaller, fine-tuned models with their more robust counterparts, developers can optimize performance and cost-effectiveness. This approach is particularly beneficial for AI developers building applications that require dynamic model selection based on specific queries.

[00:25] Host: "We use smaller models when people query our AI chat, we use smaller models to determine what model to spit their query out to next. It's way more cost effective than just using something like GPT4O and it's just as capable."

Key Use Cases Highlighted by Mistral

Mistral's LE Ministro models are designed to excel in various scenarios, emphasizing local privacy and low latency. The host outlines several compelling use cases:

On-Device Translation: Enables real-time translation without relying on internet connectivity, ideal for travelers in remote locations.
Internetless Smart Assistants: Provides assistance in areas with poor or no internet, crucial for emergency situations.
Local Analytics: Facilitates data analysis on the device, ensuring data privacy and quick processing.
Autonomous Robotics: Allows robots to operate independently without constant internet access, applicable in manufacturing and potentially military applications.

[00:40] Host: "Our most innovative customers and partners have increasingly been asking for local privacy first inference for critical applications such as on device translation, email, Internetless smart assistants, local analytics and autonomous robotics."

Performance and Pricing Advantages

Mistral's models not only offer robust functionalities but also come with competitive pricing, making them accessible for a broad range of users:

Ministral 8B: Approximately $0.10 per million tokens (input/output), translating to about 750,000 words.
Ministral 3B: Approximately $0.04 per million tokens.

These rates present a significant cost advantage over larger models, enabling extensive use without prohibitive expenses.

[01:10] Host: "Ministral 8B is about $0.10 per million input or output tokens... Ministral 3B costs $0.04 per million output or input tokens. So that's insane."

Competitive Landscape and Industry Impact

The release of LE Ministro arrives amidst a growing trend of developing smaller, more efficient AI models optimized for edge hardware. Competitors like Google with their Gemma family, Microsoft with PI models, and Meta's Llama suite are also advancing in this domain. However, Mistral asserts that their 3B and 8B models outperform existing models like Llama Gemini and Mistral’s own previous 7B model across various AI benchmarks.

[01:25] Host: "Mistral claims that Mistral 3B and Mistral 8B are going to outperform Llama Gemini as well as their own Mistral 7B on a bunch of different AI benchmarks that they're doing."

Mistral’s Open-Source Commitment and Accessibility

Mistral emphasizes an open-source philosophy, making models like Ministral 3B and Ministral 8B available for research purposes. For commercial use, developers and companies can obtain licenses through direct contact, ensuring both accessibility and controlled deployment for business applications. Additionally, Mistral offers these models through their LA platform and is in the process of partnering with major cloud providers to broaden accessibility.

[01:40] Host: "Ministral 8B right now it's available to download like you can get this today... developers can use MinisterL3B and 8B through their Cloud platform, which is called LA platform."

Future Developments and Innovations

Mistral is not resting on its laurels. The company has recently raised $640 million, underscoring investor confidence in their trajectory. Future releases include:

Pix Trail 12B: A more advanced model set to amplify Mistral’s capabilities.
Code Stroll: Their inaugural generative AI model tailored specifically for coding, expanding their application spectrum.

[02:00] Host: "They also announced their pix trail 12b. Just a lot of impressive stuff. And of course this year they also announced Code Stroll, which is their first generative AI model for code specifically."

Strategic Positioning and Market Impact

Mistral’s strategic focus on Europe, particularly France, positions the company as a key European contender in the global AI market. Their approach challenges prevailing assumptions about the dominance of larger AI companies, showcasing that innovation can thrive outside traditional hubs.

[02:15] Host: "This is definitely a company to watch, one that has raised a ton of money and is really trying to put I think Europe, but specifically France on the map with what they're able to actually produce."

Conclusion and Future Outlook

The AI Podcast host expresses strong enthusiasm for Mistral’s advancements, highlighting the company’s potential to reshape the AI landscape with its efficient, cost-effective, and versatile models. As Mistral continues to innovate and expand its offerings, it stands out as a formidable force in the AI industry, warranting close attention from developers, businesses, and AI enthusiasts alike.

[02:30] Host: "Very innovative company. [...] Mistral definitely is that company. It's the one. I'll keep you updated on moving into the future."

Notable Quotes:

Introduction of LE Ministro:

[00:02] Host: "We have big news from Mistral AI as they have just released their first new model that's specifically designed to run on laptops and phones or what we're calling Edge cases."
Hybrid Model Benefit:

[00:25] Host: "We use smaller models when people query our AI chat, we use smaller models to determine what model to spit their query out to next. It's way more cost effective than just using something like GPT4O and it's just as capable."
Use Case Emphasis:

[00:40] Host: "Our most innovative customers and partners have increasingly been asking for local privacy first inference for critical applications such as on device translation, email, Internetless smart assistants, local analytics and autonomous robotics."
Pricing Insight:

[01:10] Host: "Ministral 8B is about $0.10 per million input or output tokens... Ministral 3B costs $0.04 per million output or input tokens. So that's insane."
Competitive Edge:

[01:25] Host: "Mistral claims that Mistral 3B and Mistral 8B are going to outperform Llama Gemini as well as their own Mistral 7B on a bunch of different AI benchmarks that they're doing."
Open-Source Strategy:

[01:40] Host: "Ministral 8B right now it's available to download like you can get this today... developers can use MinisterL3B and 8B through their Cloud platform, which is called LA platform."
Future Innovations:

[02:00] Host: "They also announced their pix trail 12b. Just a lot of impressive stuff. And of course this year they also announced Code Stroll, which is their first generative AI model for code specifically."
Strategic Positioning:

[02:15] Host: "This is definitely a company to watch, one that has raised a ton of money and is really trying to put I think Europe, but specifically France on the map with what they're able to actually produce."
Final Endorsement:

[02:30] Host: "Very innovative company. [...] Mistral definitely is that company. It's the one. I'll keep you updated on moving into the future."

Additional Notes:

The episode also includes recurring mentions of AI Hustle School, a community aimed at individuals looking to monetize AI tools and scale their businesses using AI-driven strategies. While prominent in the transcript, these sections were identified as promotional content and thus were summarized briefly in the introduction without detailed coverage, in line with the request to omit advertisements and non-content segments.

For listeners interested in the technical and business implications of Mistral AI’s latest offerings, this episode provides a comprehensive overview of how these new models can be leveraged across various industries, the competitive landscape, and future prospects of Mistral as a leader in AI innovation.

Loading summary

Transcript1 lines

[00:01]
A
We have big news from Mistral AI as they have just released their first new model that's specifically designed to run on laptops and phones or what we're calling Edge cases. This is a really fascinating new development from this Paris based startup. So today on the podcast going to be diving into everything that Mistral has released with this new update and some interesting new things they've been doing in the past that you might not be aware of. Why you should keep an eye on Mistral as this is going to be a major player in the AI space moving forward. Before we get into it, if you're interested in making money with AI tools, I would love to have you a member of the AI Hustle School community. This is a community where every single week I release exclusive content showing you how I'm scaling my current businesses with AI tools, how you can do it too, all of the processes, workflows and tools that I'm using and how I'm making money with different AI tools from side Hustle. So if this is something that's interesting to you with, we have an incredible community, exclusive content and teach you how to grow your business. Join the AI Hustle School community and we'd love to have you involved in the process. Otherwise, let's get into what Mistral is currently doing. So all of this is very interesting. It's a new family of models that Mistral is calling LE Ministro and essentially they can be used or tuned for a bunch of different use cases. So they have obviously just like basic text generation, but they also have some other interesting things. Specifically they're saying that you should use them with some of their other, their bigger, more capable models. So you kind of use this hybrid of a smaller model that's more fine tuned with some of their better models. And some people I know are gonna be skeptical of this approach because they're like why would I do that? Like I'm just gonna use the best model. I'm just using GPT4O Preview to just get what I need done done. I just want the best model. So this is actually very, very useful for developers, myself included. I'm currently develop of different AI applications and specifically with AI Box, what I'm building right now, we use smaller models to when people query our AI chat, we use smaller models to determine what model to spit their query out to next. It's way more cost effective than just using something like GPT4O and it's just as capable. There's a ton of different tasks that developers are using Some of these smaller models for, they're faster, they're more nimble and of course, like they're mentioning here, you can run this directly on your phone and on your laptop, which a big benefit of this is specifically privacy and security. You're not sending this up to a cloud. No one can intercept it, no one can stop it. And so I think that for a lot of people in a lot of places is something that's very interesting to them. So like I mentioned, there's two models available, Ministral 3B and Ministral 8B. Both of them have a context window of 128,000 tokens. So this means that they can both take in about the length of a 50 page book. This is fantastic. This is as big as you need for most use cases. You know, 99% of users, this is going to be perfect. So overall, I'm really excited with what they've done here. They had a blog post kind of outlining everything going on here and they had an interesting quote which said our most innovative customers and partners have increasingly been asking for local privacy first inference for critical applications such as on device translation, email, Internetless smart assistants, local analytics and autonomous robotics. LE Ministro were built to provide a compute efficient and low latency solution for these scenarios. From independent hobbyists to global manufacturing teams, LES Ministro delivers for a wide variety of use cases. I'm really excited for this. I mean, they've outlined a bunch of really interesting use cases already right here on device translation. So you can imagine you're in a foreign country, maybe you don't have great cell reception, or maybe you don't have a phone plan there, or you don't have Internet. It can, it can translate right on, on your device and help you communicate with people there. So that's one amazing use case. Another one is Internetless smart assistance. So again, you have this situation where you're stuck in the mountains on a hike and you don't have great reception because you don't have Starlink hooked to your phone, which I'm sure this problem will disappear in the next five years. But today, and you break your foot and you want to know how to put a cast on it or get off the mountain, I don't know, right? You can think of emergency situations, but you can think of a lot of other useful situations where you just might not have Internet or your Internet's down and you want to ask a question. So smart assistants, that's great. Local analytics, autonomous robots. This is very fascinating to me. Running autonomous robots without connection to the Internet is a very interesting concept I think for probably manufacturing for, for all sorts of things. But also you could think of a lot of the military applications that that might have terrifying prospect. But you can also, you can see where that, where that goes. And then they also say that it's essentially built to provide compute efficient. So this is not going to require some crazy computer run. Like this is going to be right on your phone and then low latency. So these things are going to be incredibly fast. You're not sitting around waiting. They're very quick. And this is what we see with GPT4 mini as well is that it's very, very fast. So global manufacturers and a bunch of other people are using this on their blog posts. They go through the benchmarks. The thing is fairly robust as far as hitting different benchmarks and what it's actually able to accomplish. And I do have to give them a shout out because at the end of their blog post they did put a meme on there which it says 2023 and then they just say Mistral 7B and it's like all the Ninja Turtles, their kids with their Grandmaster in the middle. And so it's like okay, like last year we came up with one thing and then it says 2024. It's all the Ninja Turtles grown up and the Grandmaster is now smaller than them. But each of the Ninja Turtles is labeled Ministral 3B, Ministral 8B Pick's Trail, Ministral Small. I think they're outlining that last year they came up with something interesting and novel and this year they've really brought the fire and put out a ton of stuff. And again the thing you gotta remember and love about Mistral is that they do so much of this open source. So an absolutely fascinating company that's really, really doing a lot. Now the thing to, to I guess outline here is that Ministral 8B right now it's available to download like you can go get this today. So I will give them a kudos for that. But I will say that it is strictly for research purposes. If you want devs and companies, you know your devs and your and or your company to essentially use this, either their 8B or their 3B minstrel. And you want to do self, you know, development or deployment setups, you need to contact them for a commercial license. So you can do that. But that's going to be a contact thing. And obviously this is probably part of their how they plan on making money. So Otherwise developers can use MinisterL3B and 8B through their Cloud platform, which is called LA platform. And they also have, they have other clouds which they have partnered with in the coming weeks that they're going to get this on as well. So this is going to be interesting. I would speculate maybe Microsoft Azure or maybe Google or maybe even Amazon aws. So could be all or some of those that they would be getting this on right now. Ministral 8B is about $0.10 per million input or output tokens. So that's about 750,000 words. Very incredibly well priced. And their Ministral 3B costs $0.04 per million output or input tokens. So that's insane. For 10 cents you can generate a quarter million words or on their smaller model, 4 cents can generate a quarter million words. That is or sorry, not a quarter million, three quarters of a million words. So that's absolutely insane right now. They're not the only one doing this. There's definitely been a trend towards smaller models that are cheaper and quicker to train or fine tuned. I outlined a bunch of use cases, but really some of this stuff can get quite cheap. Google is adding a bunch of new models to their Gemma family. Microsoft has PI in their collections of models and recently the Llama suite introduced a bunch of new models that are really optimized for Edge hardware as well. So a lot of people are coming out with this stuff. Mistral claims that Mistral 3B and Mistral 8B are going to, you know, outperform Llama Gemini as well as their own Mistral 7B on a bunch of different AI benchmarks that they're doing. So to me that's very interesting. All in all, very impressed with this company. You know, we've recently seen them raise $640 million. They even launched a free tier for developers to test their AI models. And they've also launched a bunch of new services and an SDK to essentially let their customers fine tune their models. So a lot of impressive stuff is coming out of Mistral. Very innovative company. They've announced their pix trail 12b. Just a lot of impressive stuff. And of course this year they also announced Code Stroll, which is their first generative AI model for code specifically. So a lot of innovation coming out of AI. This is definitely a company to watch, one that has raised a ton of money and is really trying to put I think Europe, but specifically France on the map with what they're able to actually produce and you know, prove all the naysayers wrong. So I love a company with a chip on its shoulder. Mistral definitely is that company. It's the one. I'll keep you updated on moving into the future. Again, if you're interested in scaling and growing your business with AI tools, I would love for you to join the hustle school community. Or if you're trying to start an AI side hustle and make money on the side. I'd love to cover how I was able to make over $12,000 in the in a number of months, uploading free videos to Amazon and a ton of other side hustles that I've tried over the last couple years. What is currently making money for me, what's working, all the tools I use, all of that is in the AI Hustle school community. So if that's interesting, check out the link in the description and I will see you.