Mistral, the Paris-based AI startup that gained recognition for raising Europe’s largest-ever seed round, has made its mark in the programming and development industry with the launch of Codestral. Codestral is a 22B parameter, open-weight generative AI model that specializes in coding tasks. It covers over 80 programming languages, making it a valuable tool for software developers seeking to design advanced AI applications.
Enhancing Coding Workflows and Reducing Errors
Codestral’s capabilities extend beyond typical code-centric models. With a context length of 32K, developers can write and interact with code across various coding environments and projects. The model has been trained on a dataset of more than 80 programming languages, including popular ones like SQL, Python, Java, C, and C++. Additionally, it covers more specific languages like Swift and Fortran.
Mistral claims that Codestral outperforms existing code-centric models on most programming languages. It has demonstrated superior performance on evaluation tests such as RepoBench, HumanEval, and CruxEval. Notably, Codestral outperformed competitors on HumanEval for Python code generation and Bash, Java, and PHP code evaluation.
Prominent Tools Testing Codestral
Several popular tools for developer productivity and AI application development have already begun testing Codestral. These include LlamaIndex, LangChain, Continue.dev, Tabnine, and JetBrains. Harrison Chase, CEO and co-founder of LangChain, praised Codestral for its speed, favorable context window, and support for tool use in self-corrective code generation.
Getting Started with Codestral
Mistral offers Codestral 22B on Hugging Face under a non-production license for non-commercial purposes, testing, and research work. Developers can also access Codestral via two API endpoints: codestral.mistral.ai and api.mistral.ai. The former is designed for IDE integration, while the latter is for broader research, batch queries, and third-party application development.
Competition in the Code-centric Models Market
Mistral’s entry into the programming and development space with Codestral offers enterprise researchers another option to accelerate software development. However, it faces competition from other code-centric models in the market, including OpenAI’s Codex, which powers GitHub’s co-pilot service, and Amazon’s CodeWhisper tool. OpenAI’s ChatGPT and GPT-4 Turbo models also serve as coding tools. Other competitors include Replit, with its small AI coding models on Hugging Face, and Codenium, which recently secured $65 million in series B funding.
Don’t Miss The AI Impact Tour
In addition to the launch of Codestral, Mistral is inviting developers to attend The AI Impact Tour on June 5th in NYC. This exclusive event will provide insights into auditing AI models and ensuring fairness, optimal performance, and ethical compliance across diverse organizations. Request an invite to secure your attendance and take advantage of this incredible opportunity.
