Mistral AI Launches Mistral Small 3.1, Challenging the AI Giants
Paris-based AI startup Mistral AI has released Mistral Small 3.1, a new, open-source AI model designed to be both powerful and lightweight. The company claims its latest offering surpasses the capabilities of similar models created by industry leaders like OpenAI and Google LLC. This move is likely to intensify the competition in the race to develop advanced, yet cost-effective, large language models.
Mistral Small 3.1 stands out by processing text and images using just 24 billion parameters. This is a fraction of the size of many leading-edge models, yet it still manages to compete effectively. Mistral AI explains that, compared to its predecessor, Mistral Small 3.1 offers “improved text performance, multimodal understanding and an expanded content window of up to 128,000 tokens.”
The company also states the new model can process data at speeds of around 150 tokens per second, making it well-suited for applications needing rapid responses.
This is a demonstration of Mistral Small 3.1 capabilities.
A New Strategy for AI Development
This is an impressive technical feat, highlighting Mistral AI’s different strategic approach. While many competitors focus on throwing more computing power at newer models, Mistral AI prioritizes algorithmic improvements and training optimization. This strategy allows them to maximize performance with smaller model architectures. One of the main advantages of this is it makes AI more accessible. By building capable new models that can run on relatively modest infrastructure – such as a single RTX 4090 graphics processing unit or a Mac laptop with 32 gigabytes of random-access memory – it enables more advanced AI deployment on smaller devices in remote locations.
Mistral’s approach may prove to be more sustainable than continuously scaling up AI models, and it may ultimately force better-funded rivals to adopt a similar strategy.
Founded in 2023 by former AI researchers from Google’s DeepMind unit and Meta Platforms Inc., the French company has already established itself as the top AI company in Europe. To date, it has secured more than $1.04 billion in capital, reaching a valuation of about $6 billion. While substantial, this figure pales in comparison to OpenAI’s estimated $80 billion valuation.
A Growing Portfolio
Mistral Small 3.1 is the latest product release from the company. Last month, Mistral debuted a new model called Saba, focused on the Arabic language and culture. This was followed by the launch of Mistral OCR this month. Mistral OCR is a specialized model that uses optical character recognition to convert PDF documents into Markdown files, which makes them more accessible to large language models.
These specialized products build upon a broader portfolio of AI models. This includes the current flagship offering Mistral Large 2, a multimodal model called Pixtral, a code-generating model called Codestral, and a family of highly optimized models for edge devices known as Les Ministraux. This diverse portfolio demonstrates how Mistral AI is tailoring its innovation to market demand. The company is creating various purpose-built systems to meet growing needs instead of trying to compete head-on with OpenAI and Google.
Open Source Strategy
Mistral’s dedication to open-source is also a distinct strategic choice that sets it apart in an industry dominated by closed, proprietary models. It has, to some extent, paid off, with “several excellent reasoning models” being built on the foundations of its lightweight predecessor, Mistral Small 3. It shows that open collaboration has the potential to accelerate AI development more quickly than what any single company can achieve working independently.
By making its models open-source, Mistral also benefits from the expanded research and development capabilities the wider AI community provides. This allows it to compete with better-funded competitors. However, it’s worth noting that Mistral’s open-source strategy also creates challenges for revenue generation, forcing it to focus on providing specialized services, enterprise deployments, and unique applications that leverage its foundational technologies.
While the long-term viability of Mistral’s chosen path remains to be seen, it’s clear that Mistral Small 3.1 is a significant technical achievement. It reinforces the idea that powerful AI models can be made accessible in much smaller, more efficient packages. Mistral Small 3.1 is available via Huggingface and through Mistral’s AI application programming interface or on Google Cloud’s Vertex AI platform. It will also be available in the coming weeks via Nvidia Corp.’s NIM microservices and Microsoft Corp.’s Azure AI Foundry.
