The first LLM from Mistral AI created to help developers write code, Codestral, has been launched.
Though Mistral and Mixtral language models are well-known from the French business sponsored by Amazon, Codestral seems to be a step up. It is taught in more than eighty programming languages, which date back to 1957, including Python, Java, C, C++, JavaScript, Bash, Swift, and Fortran.
The LLM can write tasks, work on coding projects, and use a “fill-in” method with incomplete code while communicating in English. In addition, Codestral has been designed to lower the likelihood of errors and faults in code for businesses and developers that use it.
Just over a year after the business was established as a “European rival to OpenAI,” Mistral AI, with a valuation of £6 billion, is making waves in the sector.
The generative AI coding paradigm was referred to as “open weight” in an official blog post, however that might be disputed. Codestral cannot be used for any commercial purpose, but there is a clause for “development” that has a further catch: the rules forbid “any internal usage by employees in the context of the company’s business activities.”
Codestral’s Positive Benchmark Results
To CodeLlama 70B, DeepSeek Coder 33B, and Llama 3 70B, Mistral compared Codestral 22B. Even though Codestral uses fewer arguments, several benchmarks show that it performs well on Python, SQL, and other programming languages. Its 32k context window is larger as well.
Though it is still a young company in the AI space, Mistral AI has left its mark. In February, Microsoft reached an agreement with Amazon to utilize its models on the Azure platform, and the two parties also gained support from each other. Also, Mistral Large, the French startup’s most potent premium solution, is now available to Azure AI Studio clients thanks to the agreement.