This week in AI, Mistral AI released a new flagship language model: Mistral Large.
If you're unfamiliar with Mistral, they're a Paris-based startup made headlines last year for raising the largest seed round in European history of $113M. They've since gone on to raise another $415 million, valuing the company at $2 billion.
Now, in less than a year since that funding date, they've released multiple open source models that have been well received by the AI developer community.
Along with the release of their Mistral Large model, the startup also announced a new partnership with Microsoft to make the model available on Azure cloud computing platform.
Aside from OpenAI, this multi-year partnership marks Microsoft's second major deal to make LLMs commercially available on Azure.
Let's look at what Mistral Large is and how it compares to competitors like GPT-4.
Mistral Large Capabilities
As you can see below, evaluating on the benchmark MMLU (Measuring massive multitask language understanding), Mistral Large is said to beat Claude 2 and be just slightly lower than GPT-4.
Below you can see the performance against several other common sense, reasoning, and knowledge benchmarks for LLMs:
Here are a few more key capabilities of the model
- It has a 32K tokens context window
- Fluent in English, French, Spanish, German, and Italian
- Capable of function calling, this is a big one for LLMs - you can learn more about function calling for GPT-4 below:
Peter Foy
As the Chief AI Officer at Microsoft writes:
We have tested Mistral Large through the Azure AI Studio in a use case aimed at internal efficiency. The performance was comparable with state-of-the-art models with even better latency.
How to use Mistral Large
As of today, Mistral Large is available on "La Plateforme" and Azure. You can also test out the model yourself through their beta realase: le Chat.
Le Chat is a chat assistant that stands as a direct competitor to OpenAI's ChatGPT. Available in beta, Le Chat offers users a choice between different models, including Mistral Small and Mistral Large.
As Mistral AI's CEO, Arthur Mensch, highlights:
With Azure’s cutting-edge AI infrastructure, we are reaching a new milestone in our expansion propelling our innovative research and practical applications to new customers everywhere.
In summary, the Microsoft x Mistral partnership not only provides Mistral AI with a meaningful distribution channel, but also helps Microsoft move beyond solely partnering with OpenAI, which as FT highlights has brought scrutiny from regulators.
