ChatGPT has a big French competitor: The tech startup Mistral AI has just launched three paid AI plans – Mistral Large, Mistral Small, and Mistral Embedded – along with a free chatbot option, “Le Chat.”
Mistral AI is making a splashy entrance into the already-packed battle for AI hegemony, alongside more established competitors including Google Gemini, Claude, Copilot Pro, and many others. What makes it stand out? Mistral is claiming that its most powerful model, Mistral Large, is second only to ChatGPT 4 Turbo.
Plus, Le Chat is free to for all to use online, a move that may help it pick up attention from those hoping to use chatbots to help them plan their lives. But how much can the bot help you? Here’s what to know.
What Is Mistral Le Chat?
No, “Le Chat” is not a description of the internet from someone in 2010. It’s Mistral AI’s answer to the free ChatGPT chatbot.
Users can sign up with just an email address and a password in order to take a crack at talking to Le Chat. In response, the service will rapidly spit out an answer to any question you care to ask it.
It’s a useful service for getting first-draft responses to common concerns that you don’t want to Google for yourself: You might ask about a receipe for a great home-cooked meal or ask for fashion advice. The more guidance you can offer in your prompt, the better a response you’ll get, from the writing style to the word count.
To check it out, head over to chat.mistral.ai/chat
What Is Mistral Large?
Mistral Large refers to the powerful AI engine that Mistral offers to businesses and consumers who are willing to pay more for a better experience. It boasts a great latency-to-performance ratio, and the creators say they’ve defeated the challenges any generative AI engine must deal with: It has resisted biases, can be customized to fit a company’s needs, and operates quickly.
Companies hoping to adapt an AI tool for internal use might opt for a Mistral model if they want an option that makes self-deployment easy, or they need an option with “advanced modular moderation controls.” The service can be deployed in the cloud or on-premises.
Mistral also says that its Large model has “strong” programming language skills, and can learn and respond in multiple languages: English, French, Italian, German, Spanish.
The company also offer Mistral Small, an option that’s less costly than Large and works best for low-latency tasks, as well as Mistral Embedded, an embedding model that helps with semantic search & RAGs. (Embedded can only handle English)
Mistral explains the difference between its three models on its website.
Who Owns Mistral AI?
Mistral AI is a French AI startup, cofounded in April 2023 by former DeepMind researcher Arthur Mensch, former Meta employee Timothée Lacroix, and former Meta employee Guillaume Lample.
Arguably, Microsoft has some influence over the company, as well: In February 2024, Microsoft entered a multi-year deal with Mistral AI, giving the startup €15 million (or $16 million) to bring Mistral Large to its cloud computing platform Azure.
For Microsoft, it’s a quick way to hedge their OpenAI bets with another LLM that’s already established in Europe. Microsoft is far from abandoning its relationship with the San Francisco-based OpenAI, however, particularly given that its already invested a whopping $10 billion into the popular AI company.
Also, not that anyone was asking, but here’s the origin for the name itself: The word “mistral” is defined by Oxford Languages as a “strong, cold northwesterly wind that blows through the Rhône valley and southern France into the Mediterranean, mainly in winter.”
Mistral Tested: Is It Any Good?
We ran a few prompts through Mistral’s Le Chat, powered by the Large model. We even checked our results against ChatGPT’s responses to the exact same prompts. We’re happy to announce that Mistral offers advice that’s on par with what you’d receive from ChatGPT, from interior design advice to shopping suggestions.
Here’s a quick look at what we asked:
Click the previews to see larger versions of each image.
How Much Does Mistral AI Cost?
Here are the prices that Mistral makes public on its website for use of its LLMs:
- Mistral Large costs $8/1M tokens (input) and $24/1M tokens (output)
- Mistral Small costs $2/1M tokens (input) and $6/1M tokens (output)
- Mistral Embed costs $0.1/1M tokens (input only)
The most you can expect to pay is for the Mistral Large model, which would be $8/1M tokens (input) and $24/1M tokens (output). For context, ChatGPT 4 Turbo starts at $10/1M tokens (input) and $60/1M tokens (output), making it a noteably pricier option. It’s certainly possible that Mistral is attempting to undercut its competition.
Mistral vs ChatGPT: What’s the Difference?
Officially, there’s one big reason why ChatGPT is better than Mistral: ChatGPT 4 Turbo earns a 86.4% for its Massive Multi-Task Language Understanding (MMLU) benchmark, compared to Mistral’s 81.2% benchmark. However, ChatGPT does not offer self-deployment for businesses, while Mistral does, and Mistral offers “advanced” moderation features, compared to ChatGPT’s standard offering.
What’s the difference when using the free versions of ChatGPT and the new Mistral AI? You might be surprised at how little a difference you can notice. They’ll both respond within seconds, deliver a lengthy stream of freshly-generated text — and they’ll even often give very similar responses to the same prompt.
One question that we asked both chatbots: “Can you recommend any fashion tips for a 32-year-old man in 2024?” In both cases, ChatGPT and Mistral Large both responded with a numbered list, and the first tip from both bots had the same headline: “1. Invest in Quality Basics.”
From there, the actual text differed. Mistral explained that “Investing in quality pieces can ensure they last longer and look better,” while ChatGPT said “Build your wardrobe around high-quality basics that never go out of style.”
Critics have pointed out some differences in outputs, with some saying that Mistral offers more concise answers. One redditor said that even an earlier, open source version of a Mistral model from December 2023 was, in their opinion, better than ChatGPT.
Mistral Large vs Mistral 7b
Hugging Face is a French-American machine learning company that develops computer tools for custom applications. It was one of a few companies that received early access to Mistral AI models last year. Specifically, Hugging Face used Mistral 7B, one of two open AI models that Mistral offers.
Mistral 7B is the earliest open model, with a speedy 7B transformer that could be customized quickly, though it used a limited 8k token context window. The other open model, Mistral 8x7B, is an evolution that uses a 32k context window and offers the multi-language fluency in English, French, Italian, German, and Spanish, (as well as the programing code strengths) that you’ll get from Mistral Large and Small.
These models are available for free under a permissive license, but unless you’re heavily into open-source code and figuring out artificial intelligence, you’ll likely prefer to stop with Mistral Large or Le Chat.
