Recently, Mistral AI launched two new AI models: Codestral Mamba 7B, designed for programmers and developers for code generation, and Mathstral 7B, specifically designed for mathematical reasoning and scientific discovery.

Codestral Mamba 7B features faster inference speeds and a longer context, providing quick response times even with longer input texts. Additionally, this model can handle inputs of up to 256,000 tokens, which is double that of GPT-4o.

Mathstral 7B has a 32K context window and will be released under the Apache 2.0 open-source license. It achieves superior performance compared to other mathematical reasoning models based on benchmarks that require more inference time, and it also has fine-tuning capabilities.

1. Code Generation Model Capable of Handling Longer Contexts

The well-funded French AI startup Mistral AI is known for its powerful open-source AI models. Now, it has introduced two new entries in its growing series of large language models (LLMs): one based on mathematics and the other a code generation model aimed at programmers and developers, built on a new architecture called Mamba developed by other researchers at the end of last year.

Mamba aims to enhance the efficiency of the transformer architecture used by most leading LLMs by simplifying its attention mechanism. Models based on Mamba differ from more common transformer-based models, potentially offering faster inference speeds and larger context windows. Other companies and developers, including AI21, have already released new AI models based on it.

With this new architecture, Mistral AI aptly named its model Codestral Mamba 7B, which provides quick response times even with longer input texts. Codestral Mamba is suitable for code productivity use cases, particularly for more localized coding projects.

Mistral AI has tested the model, which will be available for free on Mistral AI's la Plateforme API, capable of processing inputs of up to 256,000 tokens, double that of OpenAI's GPT-4o.

Mistral AI has indicated that in benchmark tests like HumanEval, Codestral Mamba outperformed competitor open-source models such as CodeLlama 7B, CodeGemma-1.17B, and DeepSeek.

Developers can modify and deploy Codestral Mamba from its GitHub repository and HuggingFace. It will be provided under the open-source Apache 2.0 license.

Mistral AI claims that early versions of Codestral outperform other code generators, such as CodeLlama 70B and DeepSeek Coder 33B.

Code generation and coding assistants have become widely used applications of AI models, with platforms like GitHub Copilot powered by OpenAI, Amazon's CodeWhisperer, and Codenium gaining increasing popularity.

2. Exceptional Capabilities of the Mathematical Reasoning Model with Fine-Tuning Features

The second model launched by Mistral AI is Mathstral 7B, an AI model specifically designed for mathematical reasoning and scientific discovery. Mistral AI developed Mathstral through Project Numina.

Mathstral features a 32K context window and will be released under the Apache 2.0 open-source license. Mistral AI claims that this model outperforms all other models designed for mathematical reasoning. It achieves "significantly better results" on benchmarks that require more inference time. Users can utilize it as is or fine-tune the model.

In a blog post, Mistral AI stated, "Mathstral is another example of achieving outstanding performance when building models for specific purposes—this is a development philosophy we actively promote in la Plateforme, especially with its new fine-tuning capabilities."

Mathstral can be accessed through Mistral AI's la Plateforme and HuggingFace.

Mistral AI tends to offer its models on open-source systems, as the company has been competing with other AI developers like OpenAI and Anthropic.

Recently, the company raised $640 million in Series B funding, with a valuation close to $6 billion. It has also received investments from tech giants such as Microsoft and IBM.

The Battle for Large Model Performance Reaches New Heights

From an industry perspective, Mistral AI's new models highlight the trend of AI tools becoming more specialized. By providing powerful and accessible models like Mistral 7B and Codestral Mamba 7B, Mistral AI is becoming a significant player in the AI field, promoting innovation and the development of practical applications.

These models also emphasize the importance of open-source AI, encouraging collaboration and greater transparency within the tech community. By offering powerful AI tools to a broader audience, the rapid iteration and development of large AI models are further advanced.