French startup Mistral AI has released its first large language model for coding. Codestral gives developers looking for a code-native AI tool an option to Meta’s Code Llama, Microsoft’s GitHub Copilot and Amazon Q. Fluent in 80 programming languages — including Python, C++ and JavaScript — Codestral can complete code, write tests, and augment partial code “using a fill-in-the-middle mechanism,” while reducing “the risk of errors and bugs,” according to the company. The new LLM is described as open, but its license prohibits commercial use of both Codestral and its outputs.

“There’s a carve-out for ‘development,’” notes TechCrunch, “but even that has caveats: The license goes on to explicitly ban ‘any internal usage by employees in the context of the company’s business activities.’” (IBM’s Granite has thus far been declared the most open of generative code models.)

TechCrunch calls Mistral’s terms for its code assistant restrictive, but in a blog post about its new AI Non-Production License, Mistral says it “strikes the right balance between our commitment to openness principles and our responsibility to grow our business.”

“At 22 billion parameters, the model requires a beefy PC in order to run,” is another concern voiced by TechCrunch.

On the plus side, “the company claims Codestral already outperforms previous models designed for coding tasks, including Code Llama 70B and DeepSeek Coder 33B, and is being utilized by several industry partners, including JetBrains, Sourcegraph and LlamaIndex,” writes VentureBeat.

Some benchmark results are provided in Mistral’s Codestral announcement.

According to ReadWrite, Codestral is able to best those competitors deploying “fewer parameters” but “a greater 32k context window.”

In addition to gaining support from Amazon and Microsoft, which earlier this year agreed to use Mistral models on the AWS and Azure clouds, the company has also secured financial backing from Andreessen Horowitz, Lightspeed Venture Partners, Databricks and former Google CEO Eric Schmidt.

ITPro delves into the licensing aspect more deeply, and writes that in general, research on AI code assistants show that “developers are keen on these tools, with a study from GitHub last year revealing that users can complete tasks 55 percent faster.”

“If you want a have a closer look at Codestral, navigate to Le Chat” and after signing in, “under the prompt box, press the ‘model’ drop down and select Codestral,” ReadWrite advises.