Harnessing the Power of Mistral Large: Exploring the Latest Generative Large Language Model

by Duncan Miller on March 2, 2024
The artificial intelligence landscape has experienced yet another big development with the release of Mistral's new model of Au Large. The new large language model (LLM) from Mistral adds yet another competitive option for businesses and developers to engage with multilingual text generation, understanding, and code creation tasks. In this article, I'll explore the capabilities of Mistral Large and its implications for the industry.

Shiro Welcomes Mistral Large

We've already added Mistral large to the Shiro platform providing access to all currently available Mistral AI models including Mistral Large, Medium, Small and Tiny. Shiro enables teams to test out their prompts against Mistral's models, with direct comparison to any other model provider we offer (OpenAI, Gemini, Anthropic, Cohere) on both quantitative and qualitative metrics. If you are interested in learning more about Shiro, request a personalized demo today.

Mistral Large Features

Mistral Large is the new flagship model for Mistral AI with top-tier reasoning capabilities that position it as a contender, second only to GPT-4 in large language models accessible through an API. This model's excellence in handling complex reasoning tasks across multiple languages is competitive with the top LLMs.

Comparison on MMLU (Measuring massive multitask language understanding)


Multilingual Mastery and Precision

One of Mistral Large's key features is its fluency in English, French, Spanish, German, and Italian. This proficiency extends beyond simple translation, encompassing a deep understanding of grammar and cultural nuances, thereby enabling more contextually aware and sensitive AI applications.

Reasoning and knowledge

According to Mistral, the Mistral Large model shows powerful reasoning capabilities. In the following figure, Mistral reports the performance of the pre-trained models on standard benchmarks.

Performance on widespread common sense, reasoning and knowledge benchmarks of the top-leading LLM models on the market: MMLU (Measuring massive multitask language in understanding), HellaSwag (10-shot), Wino Grande (5-shot), Arc Challenge (5-shot), Arc Challenge (25-shot), TriviaQA (5-shot) and TruthfulQA.

Pricing

According to Mistral's pricing page, the Large model will cost $8 / 1M tokens for input and 24$ / 1M tokens for output. Compared to OpenAI pricing on the GPT-4-32K model, of $0.06 / 1K tokens for input and  $0.12 / 1K tokens for output. To compare with equal metrics, the Mistral Large model costs $0.008 / 1K tokens for input and $0.024 / 1K tokens for output. The gpt-3.5-turbo-instruct by OpenAI costs $0.0015 / 1K tokens for input and $0.0020 / 1K tokens for output.

Strategic Partnership with Microsoft Azure

Alongside the Mistral announcement, Microsoft has announced a new multiyear partnership with Mistral. The Financial Times reports that the partnership will include Microsoft taking a minor stake in the 10-month-old AI company, just a little over a year after Microsoft invested more than $10 billion into its OpenAI partnership.

Notably, Mistral has not released the weights of the Mistral Large model as they have in the past with other models like Mistral 7B and Mixtral 8X7B. This combined with the Microsoft partnership indicates a move towards closed-source development and a focus on profitability, much like we saw with OpenAI's transition from non-profit to for-profit.

On the other hand, this does mean that through Azure AI Studio and Azure Machine Learning, Mistral Large is set to become more accessible, providing a seamless experience for developers and businesses already using Azure to integrate Mistral Large into their solutions.

La Plateforme: A Gateway to Innovation

Apart from Azure, Mistral Large is also available through La Plateforme provided by Mistral AI, hosted within Europe's secure infrastructure. At Shiro, we utilize Mistral AI's API to provide access to all Mistral models.

New Mistral Small Model

Alongside Mistral Large, the launch of Mistral Small caters to needs for low latency and cost-effectiveness without compromising on performance. This optimized model serves as a scalable solution for a wide range of applications and budgetary considerations.

Empowering Development with Advanced Features

Mistral Large introduces functionalities like a 32K context window, JSON format mode, and function calling, enhancing developers' ability to interact with the model and integrate it into broader workflows. These features promise to simplify the development process, allowing for more intricate and powerful applications.

The introduction of Mistral Large by Mistral is yet another important development in the AI industry. Its strategic availability through Azure, coupled with its benchmark-setting performance, positions it as another essential tool in the toolkit for developers and businesses aiming to leverage AI's full potential. I am excited to welcome Mistral Large to the Shiro platform.

  • Photo of Duncan Miller

    Duncan Miller

    Founder, Software Developer

    Duncan is the founder and lead software developer for OpenShiro. He been running startups since 2006 and has been writing code for over 20 years. Duncan has an MBA from Babson College and lives with his wife and two children in Portland Oregon on an extinct cinder code volcano. He is passionate about artificial intelligence, climate solutions, public benefit companies and social entrepreneurship.

Subscribe to our newsletter

The latest prompt engineering best practices and resources, sent to your inbox weekly.