Mistral AI has released Codestral, the company’s first code model, designed for code generation.
Codestral is an open-weight generative AI (GenAI) model designed for code generation tasks. Open-weight models allow developers to build upon and adapt previous work, broadening AI tools’ availability to small companies, according to the National Telecommunications and Information Administration.
Codestral is trained on more than 80 programming languages, including some of the most popular ones, such as Python, Java, C, C++, JavaScript and Bash.
Codestral Capabilities
Mistral displayed test results that showed Codestral outperforms other models in various benchmarks for Python, SQL and other languages. It can complete coding functions, write tests, and complete any partial code using a fill-in-the-middle mechanism. Interacting with Codestral will help level up the developer’s coding game and reduce the risk of errors and bugs, the company said.
Codestral also performs well on more specific programming languages like Swift and Fortran. This broad language base ensures that Codestral can assist developers in various coding environments and projects.
“As a researcher at the company that created the first developer-focused GenAI tool, I’ve had the pleasure of integrating Mistral’s new code model into our chat product. I am thoroughly impressed by its performance,” said Meital Zilberstein, R&D lead at Tabnine, in a statement. “Despite its relatively compact size, it delivers results on par with much larger models we offer to customers. We tested several key features, including code generation, test generation, documentation, onboarding processes, and more. In each case, the model exceeded our expectations.”
Codestral is available for download on Hugging Face and can be used via a dedicated endpoint (codestral.mistral.ai) or the usual API endpoint (api.mistral.ai).
The model is integrated with application frameworks like LlamaIndex and LangChain, as well as IDE plugins for VS Code and JetBrains by Continue.dev and Tabnine.
“A public autocomplete model with this combination of speed and quality hadn’t existed before, and it’s going to be a phase shift for developers everywhere,” said Nate Sesti, CTO and co-founder of Continue.dev, in a statement.
New Non-Production License
Codestral is licensed under the company’s newly introduced Non-Production License to balance openness and business growth.
“Openness in AI is threatened, with the debate around it being instrumentalized to entrench the position of incumbent players in this highly competitive industry,”
the Mistral AI team wrote in a blog post. “We have raised our voices to defend openness in AI and will relentlessly continue to do so.”
Moreover, “We are therefore happy to see our community and partners build high-margin products with our models. Although this is great news for end-users, it sometimes fails to contribute to our success, research, and independence,” the team wrote. “This is why we are introducing the Mistral AI non-production license (MNPL). This license allows developers to use our technology for non-commercial purposes and to support research work. It ensures that those who build a business based on our work do so in a fair and sustainable way for all parties.”
However, Mistral will continue models and code under Apache 2.0 as the company progressively consolidates two families of products released under Apache 2.0 and the MNPL.
Steppingstone to CodeGen for the Masses
Meanwhile, Codestral is seen as a stepping stone towards empowering everyone with code generation and understanding, the Mistral AI team indicated.
Codestral is the latest salvo in the long-running battle for AI-created application code tooling, said Jason Bloomberg, an analyst at Intellyx. Today, such tools are in the hands of professional developers who can properly prompt the AI and then evaluate the quality of the resulting code, he said.
However, “Putting such powerful tools in the hands of non-technical users, however, won’t have these advantages, potentially leading to poor quality code or code that may have intrinsic quality but nevertheless doesn’t align with its business purpose,” Bloomberg told The New Stack. “Longer term, AI-generated code will ostensibly reduce the need for professional developers, leading to a shift away from the profession and thus a loss in ability to evaluate AI-generated code overall – to everyone’s detriment.”
At least one other industry analyst agreed.
“Another day another AI model, today it is Mistral with Codestral. The speed of innovation of new AI models for coding is so high, that a developer who wants to use them will spend all their time to switch to the currently best model,” Holger Mueller, an analyst at Constellation Research, told The New Stack. “But this is why all large AI vendors need to have one — exploiting the stickiness of once using a model, to keep using it. The prize all AI vendors are looking for is the low-code/no-code space — whoever gets that code generation solved in a good enough way, will start the next era of business automation.”
The post Codestral: A Step Closer to AI-Driven Coding for the Masses appeared first on The New Stack.
Codestral is seen as a steppingstone towards empowering everyone with code generation and understanding, It is the latest salvo in the long-running battle for AI-created application code tooling.