Mistral’s new Codestral code completion model races up third-party charts

Key Takeaways

• Mistral has launched Codestral 25.01, an upgraded open-source coding model, boasting twice the speed and improved performance over its predecessor.
• This new model, optimized for low-latency tasks, excels in code correction, test generation, and fill-in-the-middle tasks.
• Benchmark tests show Codestral 25.01 outperforms previous versions and other models in Python coding and achieves a high score on the HumanEval test.
• It is available to developers through IDE plugin partners, local deployment via Continue, and APIs like Mistral’s la Plateforme and Google Vertex AI, with upcoming availability on Azure AI Foundry and Amazon Bedrock.
• The release of Codestral 25.01 highlights the growing trend of specialized coding models that often surpass general-purpose models in coding tasks, reflecting a demand for specificity in development tools.