Mistral’s new Codestral code completion model is racing up the third-party leaderboards

Mistral updated its Codestral open-source coding model — which has proven popular with developers — widening the competition for coding-focused models aimed at developers.

In a blog postthe company said it upgraded the model with a more efficient architecture to create the Codestral 25.01, a model that Mistral promises will be “the clear coding leader in its weight class” and twice as fast as the previous version.

Like the original Codestral, Codestral 25.01 is optimized for low-latency, high-frequency operations and supports code patching, test generation, and mid-fill tasks. The company said it could be useful for enterprises with more data and residence model use cases.

Benchmark tests showed that Codestral 25.01 performed better on Python coding tests and scored 86.6% on the HumanEval test. It beat the previous version of Codestral, Codellama 70B Instruct and DeepSeek Coder 33B instruct.

This release of Codestral will be available to developers who are part of Mistral’s partners for IDE plugins. Users can deploy Codestral 25.01 locally through the coding assistant Continue. They also have access to the model API through Mistral’s la Plateforme and Google Vertex AI. The model is available in preview on Azure AI Foundry and coming soon to Amazon Bedrock.

More and more coding models

Mistral launched Codestral in May last year as its first code-focused model. The 22B parametric model can code in 80 different languages ​​and outperforms other code-oriented models. Mistral since then released Codestral-Mambaa code generation model built on the Mamba architecture that can generate longer code strings and handle more inputs.

And there seems to be a lot of interest in Codestral 25.01 already. Just a few hours after Mistral made its announcement, the model is already racing up the Copilot Arena leaderboards.

Writing code was one of the earliest features of base models, even for more general purpose models such as OpenAI’s O3 and Claude of Anthropic. Over the past year, however, coding-specific models have improved and often outperform larger models.

In the last year alone, there have been several coding-specific models made available to developers. Alibaba released Qwen2.5-Encoder in November. of China Coder DeepSeek became the first model to beat the GPT-4 Turbo in June. So does Microsoft revealed GRIN-MoEa mixed model based expert (MOE) that can encode and solve mathematical problems.

No one has resolved the eternal debate of choosing a general-purpose model that learns everything or a focused model that only knows how to code. Some developers prefer the breadth of options they find in a model like Claude, but the proliferation of coding models shows a demand for specificity. Since Codestral is trained to code data, it will of course be better at coding tasks rather than writing emails.

Daily insight into business use cases with VB Daily

If you want to impress your boss, VB Daily has you covered. We give you the inside scoop on what companies are doing with generative AI, from regulatory changes to practical implementation, so you can share insights for maximum ROI.

Read ours Privacy Policy

Thanks for subscribing. See more VB newsletters here.

An error occurred.