Mistral AI Unveils Codestral
A top LLM startup releases its first code-focused model!
Created on May 29|Last edited on May 29
Comment
Mistral AI has announced the release of Codestral, their first-ever generative AI model for code generation. Designed to assist developers in writing and interacting with code, Codestral supports over 80 programming languages, making it versatile for various coding environments.
A Multilingual Coding Assistant
Codestral's training encompasses a diverse dataset, including popular languages like Python, Java, C, and JavaScript, as well as more specialized ones such as Swift and Fortran. This extensive range ensures it can support a wide array of projects, providing code completions, test writing, and partial code fills.
Performance and Efficiency
As a 22-billion parameter model, Codestral offers superior performance and latency for code generation. It boasts a context window of 32,000 tokens, significantly larger than its competitors, making it highly effective in long-range code completion tasks.
Benchmark Achievements
Codestral outperforms existing models on various benchmarks. It excelled in Python code generation, SQL queries, and HumanEval tests across multiple languages, showcasing its robust capabilities. Its fill-in-the-middle performance also stands out, making it a strong competitor against models like DeepSeek Coder 33B.
Availability and Integration
Codestral is available for download under the Mistral AI Non-Production License on HuggingFace. Developers can use it via a dedicated API endpoint at codestral.mistral.ai, which is free during an eight-week beta period. For broader application development, it is also accessible on the api.mistral.ai endpoint.
Community and Tools
Mistral AI has partnered with tools like LlamaIndex, LangChain, VSCode, and JetBrains to integrate Codestral into popular development environments. This integration enhances developer productivity and enables seamless code generation and interaction.
Early adopters, including Continue.dev, JetBrains, Tabnine, Sourcegraph, LlamaIndex, and LangChain, have praised Codestral for its speed, accuracy, and efficiency. They highlight its significant impact on their development processes and its potential to transform coding workflows.
For more information on how to get started with Codestral and explore its various integrations, developers can refer to the detailed documentation provided by Mistral AI.
Add a comment
Tags: ML News
Iterate on AI agents and models faster. Try Weights & Biases today.