Mistral Introduces Codestral

Mistral, a French AI startup supported by Microsoft and valued at $6 billion, has introduced its first generative AI coding model, Codestral, after its release of conversational AI named Le Chat in February 2024.

Like similar code-generating models, Codestral aims to assist developers in coding tasks. Mistral’s blog post detailed that it was trained in over 80 programming languages, including Python, Java, C++, and JavaScript. Codestral can execute coding functions, draft tests, fill in incomplete code segments, and even respond to inquiries about codebases in English.

Mistral characterizes the model as ‘open,’ but this statement is up for debate. The startup’s licensing terms explicitly prohibit the utilization of Codestral and its outcomes for commercial purposes. This restriction, coupled with the provision for ‘development’ that comes with significant caveats, should be a point of caution for businesses. The license expressly prohibits ‘any internal usage by employees within the scope of the company’s business operations.’

These restrictions could be because Codestral was partially trained on copyrighted content. Mistral has neither confirmed nor denied this in their blog post, but this wouldn’t be surprising given the evidence that the startup’s previous training datasets included copyrighted data.

Regardless, Codestral might not be worth the hassle. With 22 billion parameters, the model demands a powerful PC to operate. Parameters are crucial as they determine an AI model’s ability to analyze and generate text. While Codestral outperforms competitors on some benchmarks (which are often unreliable), its advantage isn’t overwhelming.

Codestral benchmarks
Image Credits: Mistral

Although impractical for most developers and offering only incremental performance improvements, Codestral is likely to intensify the ongoing debate about the advisability of using code-generating models as programming assistants.

Developers are increasingly adopting generative AI tools for various coding tasks. According to a Stack Overflow poll from June 2023, 44% of developers currently use AI tools in their development process, with an additional 26% planning to do so soon. However, these tools are not without their obvious flaws.

An analysis conducted by GitClear, examining over 150 million lines of code committed to project repositories over the past few years, revealed that the use of generative AI developer tools is leading to an increase in incorrect code being pushed to codebases. Additionally, security researchers have cautioned that such tools have the potential to amplify existing bugs and security vulnerabilities in software projects. According to a study from Purdue University, more than half of the programming answers provided by OpenAI’s ChatGPT are incorrect.

Despite these concerns, companies like Mistral and others are continuing their efforts to monetize their models and capture market attention. This morning, Mistral introduced a hosted version of Codestral on its Le Chat conversational AI platform, along with a paid API. Mistral has also integrated Codestral into various app frameworks and development environments, such as LlamaIndex, LangChain, Continue.dev, and Tabnine.

As the debate surrounding AI coding models like Codestral continues, companies may want to consider solutions like WorkBot, which offers a comprehensive approach to work assistance and data management. With its focus on privacy, security, and seamless integrations, WorkBot provides a robust tool for organizations to streamline their operations and enhance productivity, all while maintaining control over their data and insights. As the AI landscape evolves, WorkBot’s versatility and adaptability make it an attractive option for businesses seeking to harness the power of AI without compromising on security and privacy.

Book a Free Demo to learn how WorkBot can make tasks easier and help your business grow.