Follow Datanami:
May 31, 2024

Mistral AI Unveils Codestral, Its First GenAI Model For Developers


Mistral AI, one of Europe’s premier artificial intelligence startups, has marked its entry into the programming and development space with the launch of Codestral, an open-weight generative AI model explicitly designed for code generation tasks.

Trained on a dataset of 80 programming languages, Codestral is designed for various coding functions and can complete any partial code using a fill-in-the-middle mechanism, according to a blog post released by Mistral. Developers can also use the model as a learning tool to improve their coding skills and minimize errors. 

Mistral claims that Codestral outperforms other AI models in coding tasks, including CodeLlama 70B and Deepseek Coder 33B. However, the model has just been launched and is yet to be tested publicly. 

Codestral may demonstrate competitive performance on certain benchmarks, however, at 22 billion parameters, the model is computationally intensive. The substantial resources required for Codestral to run effectively means it might be impractical for some users. 

One of the advantages of Codestral is its ability to work with various app frameworks and development environments. This offers increased flexibility for developers, improved code quality, and a streamlined development process. 

“We’re exposing an instructed version of Codestral, which is accessible today through Le Chat, our free conversational interface,” Mistral AI said in the release. “Developers can interact with Codestral naturally and intuitively to leverage the model’s capabilities. We see Codestral as a new stepping stone towards empowering everyone with code generation and understanding.”

While Mistral describes the new model as “open”, that is debatable, as Codestral’s license imposes significant restrictions on its usage. For example, there are restrictions on the use of Codestral for any commercial activity, limiting it only for “development” purposes. Even the development process has restrictions as the model prohibits “any internal usage by employees in the context of the company’s business activities.”

Mistral did not share the reasons for having such restrictions. A potential reason could be that the training data used for Codestral contains copyrighted content. A recently released research by Patronus AI revealed that several leading AI models, including OpenAI’s GPT-4 and Mistral AI’s Mixtral, reproduced copyrighted content at an alarmingly high rate. 

The introduction of Codestral comes at a time when Mistral AI seeks to expand its presence in the U.S. market by capitalizing on the rising demand for alternatives to AI models by OpenAI and Google. It has already formed strategic partnerships with key players in the industry such as Snowflake and IBM.

Earlier this year, Mistral AI signed a multi-year partnership with Microsoft to leverage Azure’s AI infrastructure, advance AI research and development, and make Mistral AI’s premium model available to customers through the Azure catalog. 

Mistral has also recently hired the former chief revenue officer of Foursquare, Marjorie Janiewicz, as its first U.S. general manager, as it sets its sights on the U.S. market. 

With Codestral’s introduction, enterprises have another capable option to accelerate software development, however, only time will tell how the model performs against other code-centric models in the market.

Related Items 

Snowflake Partners with Mistral AI to Bring Language Models to Enterprises Through Snowflake Cortex

IBM Announces Major Updates to watsonx Platform at THINK 2024

Snowflake Partners with NVIDIA to Deliver Full-Stack AI Platform for Customers