Mistral AI Launches Codestral: AI Code Generation Across 80 Programming Languages


Codestral is Mistral AI’s first code-focused generative AI model. It’s designed to help developers with code generation and interaction through a shared API and supports over 80 programming languages, including Python, Java, C, C++, JavaScript, Bash, Swift, and even the old Fortran. It is really good at completing functions, generating tests, and seamlessly filling in partial code, effectively minimizing errors and bugs for developers and their organizations.

Performance Benchmark and Analysis

Codestral’s ability to generate code sets a new high standard for speed and efficiency, outperforming older models designed for similar tasks.

As a 22B model, Codestral sets a new standard for code generation performance and latency compared to previous models. Detailed benchmarks show its superiority, especially in RepoBench, where its larger context window of 32k outperforms competitors with smaller windows. In evaluations across Python and SQL, it stands out by showing how well it works in different coding situations, proving its effectiveness.

Access Through its New Endpoint

Now, you can use Codestral via its own endpoint: codestral.mistral.ai. The API Key for this endpoint is managed individually and isn’t bound by typical organizational rate limits.

During an initial 8-week beta phase, access to this endpoint is free but requires joining a waitlist to ensure top-quality service. It’s particularly recommended for developers integrating IDE plugins or applications where users provide their own API keys.

Using Codestral across the platform

Codestral is accessible on multiple platforms, including the standard API endpoint: api.mistral.ai. This endpoint serves various purposes such as research, batch queries, and third-party application development. It uses tokens for billing and works well when users can access results without needing their own API keys.

Now chatting in it is also possible through Le Chat. It’s an instructed version of Codestral, available where developers can easily interact with Codestral to utilize its capabilities.

Utilize in Your Preferred Development Environment

Codestral teams up with community partners to smoothly integrate into widely-used developer tools, boosting productivity and enabling streamlined AI app development. For Application frameworks, now it easily works with LlamaIndex and LangChain for making applications. This integration facilitates the creation of intelligent applications with Codestral’s assistance, providing users with an easy and efficient development experience.

Moreover, it is now accessible within the VSCode and JetBrains environments through the integration with Continue.dev and Tabnine. This empowers developers to not only generate code but also engage in interactive conversations with it directly within their preferred coding environments.

Download Codestral

You can download it from HuggingFace under the Mistral AI Non-Production License, suitable for research and testing.

Similar Posts