Mistral AI: EU’s answer to OpenAI and Anthropic

The French start-up Mistral AI is taking the AI industry by storm and setting new precedents for both performance and accessibility. In just a few weeks after being founded, the company picked up a $113 million round of seed funding without even presenting a product. 

Their approach is pretty ambitious. According to the pitch paper, Mistral plans to create advanced AI models that are different because they will be more openly developed compared to the usual private models. This key to their approach is fostering a more transparent AI industry focused on inclusion and accessibility.

In just six months, the company has raised about $500 million in funding. This allows it to assemble a world-class team, hoard top-of-the-line GPUs, and release cutting-edge natural language models, making it a powerful contender in the world of AI giants. 

Let’s take a closer look at Mistral AI’s origin, its position in the industry, and some of the key models under its belt. 

Background of Mistral AI

Mistral AI, established in 2023, marks a significant French contribution to the global artificial intelligence landscape. This Paris-based startup was co-founded by Arthur Mensch and Guillaume Lample, both alumni of École Polytechnique (X2011), and Timothée Lacroix from École Normale Supérieure. 

The founders were fed up with how AI development was becoming too centralized and secretive. They wanted a different approach, focusing on openness, decentralization, and community involvement.

Seeing the potential for AI to change how people and machines interact, Mensch, Lacroix, and Lample saw a chance to shape its development in line with European values, making sure it was ethical and responsible.

The founders have strong AI backgrounds, having worked at industry leaders like Meta and Google’s DeepMind. They’re experts in different areas of AI, especially in natural language processing (NLP) and machine learning (ML), and have published many scholarly papers.

Mistral AI’s mission revolves around the development of open-source artificial intelligence models. If this self-hosted, open-source, and non-demanding approach takes off, we might be seeing hundreds, if not thousands, of use cases pop up instantly. Small businesses can use open-source AI models to ideate marketing campaigns, enterprises can use it for Workday staff augmentation, and freelancers can gain their own personal assistants overnight

This approach distinguishes Mistral AI from many competitors, underlining its commitment to transparency and community-driven advancement in AI.

The funding milestone

Initially kick-starting its journey with an impressive seed funding of €105 million, the company quickly set a strong foundation, showcasing its potential for innovation and growth in the artificial intelligence domain.

The watershed moment came in October 2023, when Mistral AI secured an additional €385 million, boosting its valuation to around $2 billion. This remarkable investment reflects deep confidence from investors in Mistral AI’s vision, technological prowess, and potential to stand toe-to-toe with global AI leaders such as OpenAI, Google’s DeepMind, and Anthropic.

Some of Mistral’s key investors include:

• Lightspeed Venture Partners
• Andreessen Horowitz
• Redpoint
• Index Ventures
• BNP Paribas
• First Minute Capital
• LocalGlobe

The funds from this round are earmarked for critical areas such as research and development, team expansion, and product enhancement. 

This financial boost is instrumental for Mistral AI, enabling it to scale its operations, advance its AI models, and extend its market reach. It signifies a crucial phase in Mistral AI’s growth, positioning it as a key player in the competitive AI landscape and highlighting the burgeoning interest in European AI startups.

This funding milestone is a vote of confidence in its future potential. It underscores the company’s commitment to pushing the boundaries of AI technology, making advanced AI models more accessible, and fostering innovation in the global tech ecosystem.

Mistral’s key pioneering AI models

Mistral has already introduced several key AI models that represent significant advancements in the field of artificial intelligence. Here are some of the models that are already making headlines:

Mistral 7B

The Mistral 7B model is a generative AI platform characterized by its 7.3 billion parameters. It’s designed to outperform larger models on various benchmarks and offers a compelling balance between size and capability. Key features include:

• Superior performance to Llama 2 13B across all benchmarks and comparable performance to Llama 1 34B on many benchmarks.
• Excellent at coding tasks while maintaining strong English language performance.
• Utilizes Grouped-query attention (GQA) for faster inference and Sliding Window Attention (SWA) to manage longer sequences more efficiently.
• Available under the Apache 2.0 license, allowing unrestricted use, including deployment on cloud platforms and availability on HuggingFace for easy access and fine-tuning

Mixtral 8x7B

The Mixtral 8x7B represents a leap forward in AI model development, employing a sparse mixture of expert (MoE) architecture to achieve high performance with efficient computation. Key features include:

• Handling of 32k-token contexts and proficiency in multiple languages (English, French, Italian, German, and Spanish) along with coding capabilities.
• Achieved an 8.3 score on the MT-Bench when fine-tuned for instruction-following tasks.
• The architecture allows for selective engagement of parameters, making both training and inference processes significantly faster compared to traditional models.
• Released checkpoints are compatible with optimization tools like Flash Attention 2 and bitsandbytes, further enhancing its performance and efficiency​.

It’s clear that Mistral AI is already making big strides. This comes at a time when OpenAI, known for developing GPT-4, has successfully integrated its technology into Microsoft’s government services, drawing attention from the European Union (EU). The EU views Mistral AI as its response to OpenAI’s achievements. 

However, suggesting that the AI industry is dominated by just Mistral AI and OpenAI (a duopoly) is premature, especially considering the presence of other notable players like Anthropic. While Mistral AI is emerging as a strong contender in AI, it’s still too early to make any assumptions.

How this European startup is challenging the giants

The company’s flagship model, Mixtral 8x7B, is already making headlines. It handles tasks in five languages and is designed to run up to six times faster than its proprietary counterparts like OpenAI’s GPT-3.5 and Meta’s LLaMA 2. 

This model, along with Mistral 7B, demonstrates Mistral AI’s commitment to pushing the boundaries of AI​​. That’s also why these AI giants are already “suddenly” all for AI regulation, seeing Mistral as a threat.

Mistral AI aims to be an open and transparent AI service, different from the usual closed-off AI technologies. It focuses on sharing its development process and encouraging community input to improve its AI models’ effectiveness and safety. 

The startup’s ambitious roadmap includes releasing a family of text-generating models that significantly outperform existing solutions like ChatGPT’s GPT-3.5 and Google Bard. Some of these models will be open-source, broadening the accessibility of AI technology to a wider audience and enabling more secure and customizable solutions for businesses​​.

For instance, imagine your very own completely enclosed and private document generation system. This can be a massive advancement in business security as a whole, as Google, Microsoft, and other tech giants will suddenly lose control over everyone’s information. 

Despite facing stiff competition from established tech giants, Mistral AI’s unique value proposition, focus on open-source development, and strategic partnerships, such as with NVIDIA, position it as a promising disruptor in the AI landscape. 

The company’s approach could pave the way for a more open, collaborative, and ethical AI development ecosystem, challenging the dominance of proprietary models and reshaping how AI advancement unfolds going forward.

Is Mistral AI the next big thing?

Mistral AI is making big moves in AI, opening up new possibilities by focusing on sharing its work and involving the community. This approach is attracting attention and funding, showing that it’s serious about making AI accessible and responsible, especially in Europe​​​​​​. 

With its commitment to following new EU rules for AI, Mistral AI is aiming to lead by example in creating AI that’s safe and respects people’s rights. As AI keeps advancing, Mistral AI’s open and community-focused strategy could spark more innovation and make powerful AI tools available to everyone. This marks a step forward in how AI is developed and used globally.