Mistral AI introduces its first LLM for coding, fluent in 80 programming languages
Mistral AI has confirmed the launch of Codestral, its first LLM designed to assist developers write code. The Amazon-backed French… Continue reading Mistral AI introduces its first LLM for coding, fluent in 80 programming languages
The post Mistral AI introduces its first LLM for coding, fluent in 80 programming languages appeared first on ReadWrite.
Mistral AI has confirmed the launch of Codestral, its first LLM designed to assist developers write code.
The Amazon-backed French company is known for its Mistral and Mixtral language models, but Codestral appears set to raise the bar higher. It is trained in over 80 programming languages including Python, Java, C, C++, JavaScript, Bash, Swift, and Fortran, which can be traced back to 1957.
Operating in English, the LLM will be able to complete coding tasks, write tasks, and use a ‘fill-in’ mechanism for partial code. The program has also been tipped to reduce the risk of errors and bugs in code, for those companies and developers who utilize Codestral.
With a valuation of £6 billion, Mistral AI is making waves in the industry just over a year since the startup was formed as a “European rival to OpenAI”
In an official blog post, the generative AI model for coding was described as “open weight” but that could be contested. The license prohibits Codestral’s use for any commercial purposes but there is a clause for “development” with a further catch as the terms rule out “any internal usage by employees in the context of the company’s business activities.”
Today we are releasing Codestral-22B, our first code model! Codestral is trained on more than 80 programming languages and outperforms the performance of previous code models, including the largest ones.
It is available on our API platform, through instruct and… pic.twitter.com/YS85u0oIfG
— Guillaume Lample @ ICLR 2024 (@GuillaumeLample) May 29, 2024
Codestral’s encouraging benchmark performance
Mistral compared Codestral 22B to CodeLlama 70B, DeepSeek Coder 33B, and Llama 3 70B. Fewer parameters are deployed by Codestral but it appeared to perform favorably on Python, SQL, and other programming languages according to various benchmarks conducted. It also has a greater 32k context window.
Mistral AI has made an impression despite its fledgling status in the AI sector. As well as gaining backing from Amazon, Microsoft agreed terms with the company in February to use its models on the Azure platform. The deal also enables Azure AI Studio customers to avail of Mistral Large, the premium offering, and the most powerful, from the French startup.
If you want a have a closer look at Codestral, navigate to Le Chat, and under the prompt box, press the ‘model’ drop down and select Codestral.
Image credit: Ideogram
The post Mistral AI introduces its first LLM for coding, fluent in 80 programming languages appeared first on ReadWrite.