In short
Codestral is a specialized 22B parameter LLM from Mistral AI designed for high-performance code generation across 80+ languages. Its Fill-in-the-Middle capability and 32k context window make it ideal for real-time IDE autocomplete and complex codebase analysis.
Frequently asked
- What is the primary innovation of Codestral?
- Its 'Fill-in-the-Middle' (FIM) capability allows it to predict missing code segments within existing files, making it highly effective for real-time IDE integration.
- How many programming languages does Codestral support?
- Codestral was trained on a dataset encompassing more than 80 programming languages, including Python, Java, C++, JavaScript, Fortran, and Swift.
- What is the context window size of Codestral?
- The model employs a large context window of 32k tokens, enabling it to process complex project structures and long-form codebases.
- What are the API pricing rates for Codestral?
- Pricing is $1.00 per 1M input tokens and $3.00 per 1M output tokens via Mistral's La Plateforme, with custom pricing available for enterprise use.
- Can Codestral be used for commercial purposes?
- Yes, but it operates under the 'Mistral AI Non-Commercial License' for self-hosted research and testing, while commercial use is available via their La Plateforme or integration partners.